Euterpe — Features

Tags: scope:euterpe · layer:domain · type:lib

Overview#

Euterpe is the comprehensive music technology platform for the Oshun ecosystem. Named after the Greek Muse of music and lyric poetry — one of the nine daughters of Zeus and Mnemosyne, whose name means "Giver of Delight" — Euterpe spans foundational pitch representation and music theory through AI-powered composition, professional studio production, immersive spatial audio, voice synthesis, guitar simulation, music education, film scoring, blockchain rights management, multi-platform distribution, marketing automation, real-time accompaniment, accessibility, and ethnomusicology. The domain is provided entirely as consumable TypeScript libraries with no application tier, designed to power Calliope's AI artists, Aphrodite's streaming platform, and any other domain in the ecosystem that needs deep music intelligence. Cross-domain content protection is provided by Themis Music Shield and Universal Originality Shield (Phase 74); @euterpe/protect performs analysis and scoring, while legal copyright adjudication is delegated to Themis.

The workspace currently contains 35 implemented libraries under libs/euterpe/ — 34 from TODO Phase 38 (sections 38.1–38.34) plus @euterpe/ai-scoring from Phase 70.6 — each a real package with per-feature sub-modules and Vitest specs. A further 10 packages and one browser DAW application are in the Phase 38 backlog (sections 38.35–38.45) and are not yet built; they are marked (planned) below. For the authoritative type-level contracts of the implemented libraries see specifications.md.

Domain Libraries#

Implemented (35)#

These libraries exist under libs/euterpe/ with real source and tests. Each has per-feature sub-modules, a types.ts, implementation files, and Vitest specs. The first seven rows (Core through Spatial) form the foundational layers described in architecture.md; the remaining 28 rows are the production and application tier.

Planned (Phase 38.35–38.45)#

These packages appear in TODO Phase 38 but have no directory in the repository yet. They represent the infrastructure tier that will sit above the current implemented libraries: provider connectivity, project persistence, generation orchestration, quality evaluation, observability, and the opinionated studio-runtime product layer. See the "Provider and Infrastructure Layer" section below for detailed descriptions of each planned package.

Music Primitives and Foundations (@euterpe/core)#

@euterpe/core is the foundation every other Euterpe library depends on. It defines the vocabulary — what a pitch is, what an interval is, what a chord is — so that all higher-level code speaks the same language. Because it has zero runtime dependencies, it runs equally well in a Node.js server, a browser, an edge runtime, or a WASM host.

Pitch and Note Representation#

Every musical calculation in Euterpe begins with precise pitch representation that correctly handles all musical traditions.

• Chromatic pitch classes: All 12 chromatic tones with full sharp, flat, double-sharp, and double-flat accidental support.
• Enharmonic equivalence: Automatic recognition that C# and Db represent the same pitch with context-sensitive spelling — the system understands that a D# is functionally different from an Eb even at the same frequency.
• Scientific pitch notation: Notes expressed as letter-octave pairs (C4 = Middle C, A4 = concert A, Bb5, etc.) across octaves -1 through 10.
• MIDI note numbers: Full MIDI 0–127 range with bidirectional conversion between MIDI numbers, frequencies, and note names.
• Frequency calculation: Precise Hz for any note at any tuning reference with cent-level microtonal offsets.
• Microtonal notes: Arbitrary cent offsets from equal temperament for quarter-tone, sixth-tone, and fully custom microtonal systems.
• Pitch bend modeling: Continuous pitch deviation from a base note for expressive portamento and vibrato representation.

Tuning Systems#

Different musical traditions and use cases require different reference frequencies and interval ratios. The table below lists every tuning system supported, along with the context in which each is typically used.

Notation Systems#

Six notation formats for global music traditions: scientific pitch (C4/D#5); Helmholtz notation; moveable-do and fixed-do solfège; Nashville number system (used in country and contemporary Christian music for transposable chart notation); Indian sargam (Sa Re Ga Ma Pa Dha Ni); and Byzantine neumes (ison, oligon, petaste, kentemata, hypsele, apostrophos, elaphron, chamele — the notation system of Eastern Orthodox liturgical music).

Intervals#

• Interval construction by quality and number (minor 3rd, major 7th, augmented 4th, diminished 5th).
• Inversion (a minor 3rd inverts to a major 6th; a major 2nd inverts to a minor 7th).
• Quality classification from raw semitone distance.
• Compound interval support (9ths, 10ths, 11ths, 13ths and beyond).
• Consonance/dissonance scoring; tritone detection and resolution suggestion; psychoacoustic roughness calculation.

Scales, Modes, and World Music Systems#

• Western diatonic: Major scale; all seven modes from Ionian through Locrian; natural minor, harmonic minor, melodic minor (with different ascending and descending forms).
• Pentatonic and blues: Major and minor pentatonic; hexatonic and heptatonic blues scales.
• Symmetric scales: Whole-tone scale; chromatic scale; diminished (half-whole and whole-half); augmented scale.
• Bebop scales: Bebop dominant, major, minor, and Dorian — the scales that result from adding chromatic passing tones to diatonic scales so chord tones fall on the beats.
• Indian ragas: Traditional ascending/descending raga patterns — the aroha and avaroha of each raga define its characteristic melodic movements and permitted note sequences.
• Arabic maqamat: The modal system of Middle Eastern, North African, and Turkish music — each maqam defined by characteristic interval patterns, starting note, predominant notes, and cadential patterns.
• Japanese scales: Hirajoshi (the characteristic koto scale), Iwato, In, Insen, and other traditional scales.
• Additional features: Mode rotation; scale degree naming (tonic, supertonic, mediant, subdominant, dominant, submediant, leading tone); scale-to-chord mapping with Roman numeral analysis; scale search and filtering.

Chords and Voicings#

Complete chord vocabulary from power chords and triads through 13th chords and polychords. Voicing strategies: close, open, drop-2, drop-3, drop-2-4, and spread voicings. Slash chords (C/E, D/F#), cluster chords, sus2/sus4, add-chord variants. Chord symbol parsing from standard notation strings (Cmaj7, Dm7b5, F#sus4/A). Full chord-to-MIDI conversion with specified voicing.

Rhythm and Time#

Simple (2/4, 3/4, 4/4), compound (6/8, 9/8, 12/8), complex (5/4, 7/8, 11/8), and additive time signatures. Note durations from whole through 128th notes with dotted, double-dotted, and arbitrary tuplet support. Polyrhythm detection and generation. Swing and shuffle quantization values. Metric modulation calculation. Rhythmic displacement utilities.

Keys and Tonality#

Key signature management for all major and minor keys; relative and parallel key detection; circle of fifths utilities; key distance calculation; common tone analysis between keys; key detection from note collections; modulation path finding through pivot chords; secondary dominant identification; borrowed chord and modal mixture detection.

Musical Form and Structure#

Structural unit hierarchy: phrase → period → section → movement. Common formal templates: sonata form (exposition, development, recapitulation), rondo (ABACA/ABACABA), binary (AB/AABB), ternary (ABA), strophic, through-composed, 12-bar blues, 32-bar AABA, verse-chorus-bridge, fugue structure.

Dynamics and Expression#

Dynamic markings ppp through fff with niente (fading to nothing); crescendo and diminuendo hairpins; all standard articulation marks (staccato, legato, accent, tenuto, marcato, sforzando); ornaments (trill, mordent, turn, grace note); tempo modifiers (ritardando, accelerando, rubato, fermata); expression marks (dolce, espressivo, cantabile, con fuoco); humanization parameters including timing variation (push/pull) and velocity curves.

Audio Format Handling#

Audio buffer creation at any sample rate; bit depth handling (16-bit integer, 24-bit integer, 32-bit float); WAV, MP3, FLAC, OGG Vorbis, AAC/M4A, AIFF format support; sample rate conversion between standard rates (22050–192000 Hz); channel configuration (mono, stereo, multichannel); audio normalization, concatenation, trimming, and slicing; ID3 and Vorbis comment metadata handling.

MIDI Protocol#

Full MIDI 1.0 message types: note on/off, control change (all 128 controllers), program change, pitch bend, channel aftertouch, polyphonic aftertouch, system exclusive (SysEx). 16-channel management. Standard MIDI File (SMF) parsing and generation for both Type 0 (single track) and Type 1 (multi-track). MIDI 2.0 protocol support. MPE (MIDI Polyphonic Expression) support for per-note pitch bend and pressure. MIDI quantization and transposition utilities.

Music Theory Intelligence (@euterpe/theory)#

@euterpe/theory builds on @euterpe/core to add music intelligence: it can analyze what a chord progression is doing, validate whether a melody line respects counterpoint rules, identify a genre from musical features, and generate exercises for ear-training. Calliope uses it to ensure AI-generated songs are stylistically coherent; the conservatory uses it to power its theory curriculum and adaptive ear-training engine.

Harmonic Analysis#

• Roman numeral analysis: Identifying chord functions within a key (I, ii, V7, etc.) and labeling secondary dominants (V/V, V/ii, V/vi).
• Functional harmony classification: Tonic (T), subdominant/pre-dominant (S/PD), and dominant (D) function identification — the backbone of functional harmonic syntax.
• Modal mixture and borrowed chords: Detecting chords borrowed from the parallel major or minor.
• Chromatic chords: Neapolitan (bII) chord detection; augmented sixth chords — Italian (It+6), French (Fr+6), and German (Ger+6) augmented sixths — the chromatic harmonies of 18th-19th century classical practice.
• Harmonic rhythm analysis: How often chords change, and whether harmonic rhythm accelerates (increasing tension) or decelerates (relaxing tension).

Counterpoint#

Species counterpoint rules and validation for all five species: note-against-note (1st species), two notes per cantus note (2nd species), four notes per cantus note (3rd species), syncopated (4th species), and florid (5th species). Voice leading validation detecting parallel fifths, octaves, and hidden fifths. Dissonance preparation and resolution checking. Cantus firmus generation.

Melody Analysis and Generation#

Contour classification (ascending, descending, arch, inverted arch, static, terraced); range and tessitura analysis; climax identification; motivic pattern and sequence detection; conjunct/disjunct motion balance; melody generation constrained by harmonic context, range, and style.

Chord Progression Analysis#

Common progression pattern catalog by genre (I-V-vi-IV in pop; ii-V-I in jazz; i-VII-VI-VII in rock; I-IV-I-V in blues); progression generation with voice leading optimization; functional harmony analysis.

Groove and Quantization#

Rhythmic quantization (straight, swing at various ratios, shuffle); groove template application; humanization algorithms adding natural timing and velocity variation; groove extraction from live performance data.

Pitch Class Set Theory#

Prime form computation; normal order calculation; Forte number lookup (the standard catalog of pitch class sets); interval vector; Z-relation identification (two sets with the same interval vector but different prime forms); set-class catalog; transformational operations (transposition, inversion, multiplication by M5 and M7).

Orchestration#

Instrument range and transposition reference data; scoring guidelines for combining instruments effectively; instrument combination suggestions; orchestral balance analysis; arrangement voicing distribution.

Ear Training#

Interval, chord quality, scale, and chord progression recognition exercise generation; difficulty progression from simple to complex; performance tracking and adaptive difficulty adjustment.

AI Music Generation (@euterpe/genesis)#

@euterpe/genesis is the AI generation layer. It depends on both core and theory so that generated output is harmonically grounded — a melody generation call knows the chord progression; a stem generation call knows the key. The infrastructure sub-module abstracts model loading and GPU/CPU dispatch so generation code never names a specific provider. The planned @euterpe/providers package (see Provider and Infrastructure Layer) will own actual provider API connectivity on top of this abstraction.

Text-to-Music#

Generates audio from natural language descriptions using diffusion-based and transformer-based models. Prompt parsing extracts musical intent — style, instrumentation, mood, tempo, key, and structural shape — and the genesis infrastructure sub-module provides a model registry, capability matching, and routing decisions across model backends. (A dedicated @euterpe/providers provider-connectivity layer is planned; see Provider and Infrastructure Layer.)

Stem Separation#

Audio stem separation into vocals, drums, bass, and other instruments using deep neural networks (Demucs and similar architectures). Enables remix workflows, selective regeneration of individual stems, and stem mastering.

Voice Generation#

Singing voice synthesis from lyrics and melody; vocal style transfer; voice-conditioned generation for specific vocal characters.

Melody Generation#

Creates melodic lines constrained by harmonic context, style parameters, and structural requirements using both rule-based and neural approaches. Full integration with @euterpe/theory for harmonically coherent melody generation.

Arrangement#

Automated arrangement creation from lead sheets or simple inputs — distributing melodic, harmonic, and rhythmic ideas across instruments with appropriate voicings, rhythms, and dynamics for the requested genre.

Audio Inpainting#

Fills selected regions of existing audio with contextually appropriate generated content, enabling seamless editing and track extension without obvious joins.

Style Transfer#

Transforms the style of existing musical content while preserving structural and melodic identity — enabling cross-genre adaptation of a piece without changing its underlying musical DNA.

Conditional Generation#

Generates music conditioned on multiple simultaneous constraints: specific chord progression, specific melody, rhythmic pattern, instrumentation requirements, and energy curve.

Quality Assessment#

Perceptual quality metrics for generated audio; musical coherence scoring; AI artifact detection (the "stuttering," "smearing," and unnatural transitions common in poorly generated AI audio).

Digital Audio Workstation (@euterpe/studio)#

@euterpe/studio is a full DAW engine in library form — the same set of capabilities a product like Yemaya exposes to its users, packaged so any application can embed them. It builds conceptually on core, theory, and genesis: the timeline and MIDI editor use core primitives; the AI features call genesis for stem generation and arrangement. The library ships no server; persistence and the UI shell are the host application's responsibility.

DAW Engine#

Core audio processing graph with real-time rendering; buffer management; precision latency compensation across all tracks; transport controls (play, pause, stop, record, loop with loop in/out points).

Timeline#

Arrangement timeline with track lanes, clip regions, marker system, tempo map (supporting tempo changes), time signature changes, and zoom/scroll navigation.

Clip Editing#

Non-destructive audio clip manipulation: cut, copy, paste, trim, fade in/out, crossfade, time stretch (tempo-independent pitch-preserving stretch), and pitch shift (pitch-independent tempo-preserving transpose).

MIDI Editing#

Piano roll editor with click-and-drag note editing; step sequencer; MIDI quantization with swing control; velocity editing; controller lane editing (drawing automation for any MIDI CC); MIDI transformation tools (transpose, reverse, randomize).

Mixer#

Channel strip architecture with gain staging, pan, mute/solo, send/return routing (for reverb and delay buses), bus grouping (for stem mixing and parallel compression), and master output with professional metering.

Effects Processing#

• EQ: Parametric EQ, graphic EQ, shelving filters; linear phase mode; spectrum analyzer overlay.
• Dynamics: Compressor (with threshold, ratio, attack, release, knee, makeup gain); limiter; noise gate; expander; transient shaper.
• Time-based effects: Reverb (algorithmic and convolution); delay (tape, analog, digital, multi-tap); chorus; flanger; phaser.
• Saturation and distortion: Tape saturation, tube saturation, hard/soft clip.

Virtual Instruments#

Synthesizers: subtractive (classic VCO/VCF/VCA architecture), FM (frequency modulation with carrier/modulator operators), wavetable (scanning through waveform tables), granular (real-time granular synthesis from any audio). Samplers with velocity layers and round-robin. Drum machines. Instrument preset management system.

Automation#

Breakpoint automation with linear, bezier, and step interpolation curves; automation lanes per track parameter; automation recording from real-time controller input (recording knob movements during playback).

AI-Assisted Production#

• Auto-mixing: AI mix engineer that analyzes all tracks and sets levels, panning, EQ, and compression based on learned production aesthetics.
• Intelligent EQ suggestions: Analyzing spectral overlap between tracks and suggesting EQ cuts to improve clarity.
• Mastering assistance: Analyzing the mix and suggesting mastering chain settings.
• AI audio repair: Noise reduction, click removal, hum elimination, de-reverb.

Audio Mastering (@euterpe/master)#

@euterpe/master handles the final stage of music production: preparing a finished mix for distribution at the correct loudness target, format, and quality level for every platform. Beyond its core mastering functions, it also owns the loudness math for V2 commentary mix ducking — the dB/linear conversion, sidechain envelope, and loudness delta that the @v2/euterpe-commentary-ducking bridge uses to lower V2.DynamicMusic under the V2.Commentary bus.

Mastering Chain#

• Multiband compression: Independent compression in frequency bands (typically 3–5 bands) for controlling dynamic range per-frequency rather than across the full spectrum.
• Stereo enhancement: Mid-side processing, stereo widening, mono compatibility checking.
• Harmonic excitation: Adding subtle high-frequency harmonics to add perceived brightness and air.
• Limiting: True-peak limiting with lookahead; inter-sample peak detection.
• Loudness normalization: LUFS targeting per streaming platform — Spotify targets -14 LUFS integrated; Apple Music -16 LUFS; YouTube -14 LUFS; CD -9 to -12 LUFS. The platform specifications are maintained and updated.

Mix Analysis#

Spectral analysis; dynamic range measurement (DR score); stereo field visualization (Goniometer/Lissajous); loudness metering (integrated, short-term, and momentary LUFS); true peak detection; crest factor analysis.

Stem Mastering#

Per-stem processing of drums, bass, vocals, and instrumental stems before final limiting — providing significantly more control than stereo-only mastering while being simpler than full mix-down access.

Format Mastering#

Sample rate conversion; bit depth dithering (TPDF and noise-shaping dither for 16-bit output); codec encoding (WAV, FLAC, AAC, MP3 at multiple bitrates, Opus); format-specific metadata embedding.

AI Mix Assistant#

AI-powered mix improvement suggestions from spectral balance analysis (comparing against reference tracks), dynamic range assessment, and genre-appropriate loudness targeting.

V2 Commentary Mix Ducking#

@euterpe/master owns the loudness delta, dB/linear gain conversion, and sidechain envelope math for V2 commentary mix ducking. The @v2/euterpe-commentary-ducking bridge uses those mastering primitives to lower V2.DynamicMusic under the V2.Commentary bus, stays off-rollback, and exports only audio-mix automation to V2Audio.

Spatial and Immersive Audio (@euterpe/spatial)#

@euterpe/spatial handles immersive audio formats — Dolby Atmos, binaural headphone rendering, ambisonics, and VR room acoustics. Unlike the other production libraries, it builds on core alone rather than on theory, because spatial audio processing is a signal-processing discipline that operates on audio buffers and does not need knowledge of harmony or scales. This keeps the library applicable in non-music contexts such as VR games or film post.

Dolby Atmos#

Object-based audio positioning — rather than channel-based mixing, each sound source is placed in 3D space and the renderer places it in the speaker array or headphone mix. Bed channel management for ambient sound fields; metadata generation; renderer integration; ADM (Audio Definition Model) file creation for interchange.

Binaural Rendering#

HRTF (Head-Related Transfer Function) processing — mathematically filtering audio to simulate the cues the outer ear, head, and body create when sound comes from different directions. HRTF database management; personalized HRTF estimation from listener photos or measurements; head tracking integration for dynamic rotation.

Ambisonics#

Higher-order ambisonics (HOA) encoding, decoding, and transformation — a scene-based spatial audio format that represents the entire soundfield mathematically using spherical harmonics. B-format (first-order, 4 channels) and higher-order spherical harmonics up to 7th order. Rotation, beamforming, and speaker-array decoding for arbitrary loudspeaker arrays.

VR/AR Audio#

Room acoustics simulation including early reflections and late reverberation; distance attenuation models; occlusion modeling (sound blocked by walls and objects); real-time source positioning tied to 3D scene state for VR and AR applications.

Spatial Upmixing#

AI conversion of stereo or multichannel audio to immersive spatial formats — intelligently distributing elements across the spatial field by identifying instrument stems and placing them in appropriate spatial positions.

Voice Technology (@euterpe/voice)#

@euterpe/voice covers the full spectrum of voice synthesis and processing — from cloning a real voice to generating singing from lyrics, from natural TTS to choir simulation. All voice cloning and conversion operations are structurally consent-gated: the ConsentRecord type is a required parameter for any operation that modifies voice identity, making consent a compile-time requirement rather than a runtime check. Iris consumes @euterpe/voice for conversational TTS; Calliope uses it to give AI artists a singing voice.

Voice Cloning#

Synthetic voice models from reference audio samples; speaker verification and consent management; reproduction for narration and singing applications.

Voice Conversion#

Speaker identity transformation preserving linguistic content — the spoken words remain the same but the voice quality changes. Encoder-decoder, GAN-based, and diffusion-based approaches; real-time and offline processing modes.

Text-to-Speech#

Multi-language natural speech generation; speaking style and emotion control; SSML markup support for fine-grained prosody control; prosody adjustment (rate, pitch, emphasis).

Text-to-Singing#

Singing vocal synthesis from lyrics and melody input; multiple vocal styles; vibrato rate and depth control; breath simulation; expression parameters for dynamics and phrasing.

Vocal Processing#

Pitch correction (both subtle transparent correction and the "auto-tune" effect for creative use); formant shifting (changing the character of a voice without changing pitch); vocal doubling (creating a thickened stereo effect); harmonizer (generating harmony parts from a melody); de-essing; breath removal; noise gate.

Choir Simulation#

Ensemble vocals from single voice inputs — synthesizing multiple virtual singers with natural variation in timing, pitch, and timbre for realistic choral textures. Configurable ensemble size, voice type distribution, and tightness.

Voice Analysis#

Real-time pitch tracking; formant analysis (identifying the resonant frequency peaks that characterize vowel quality); voice quality assessment (breathiness, roughness, strain); singing technique evaluation; vocal range detection.

AI Virtual Artist Platform (@euterpe/virtuoso)#

The music-specific layer of autonomous AI artist management. The domain boundary here is important: @euterpe/virtuoso is a companion to @calliope/* — it owns the music catalog, release pipeline, and avatar concerns while Calliope owns persona psychology, fandom, and career arc. Calliope persona, career, and fandom features are documented in DOMAINS/calliope/features.md. Built from six sub-modules: artist-identity, content-generation, avatar-system, social-engagement, performance, and business. It consumes @euterpe/genesis for music generation and @euterpe/theory for style-consistency analysis.

Artist Persona Definition#

The artist-identity sub-module models a virtual artist's creative identity: an artist persona with primary and secondary genre specs, vocal characteristics and range, visual appearance and avatar style (2d / 3d / realistic / anime / abstract), personality traits, a biography, and a ranked set of artistic influences (primary / secondary / subtle). An artist moves through an artist status of draft, active, hiatus, retired, or archived. The sub-module also generates artist names, a brand identity kit, and a discography plan.

Content Calendar Management#

The content-generation sub-module plans and produces release-promotion, behind-the-scenes, fan-interaction, and off-topic content across destination platforms, with content-type variation enforced so a platform does not receive the same type repeatedly.

Music Catalog Management#

Tracks the release pipeline at single, EP, and album granularity through the discography plan (release types single / ep / album / deluxe). Style-consistency enforcement scores candidate tracks against the artist's genre and influence profile. Originality clearance delegates copyright verification to Themis (see Domain Boundaries) and records the verdict.

Avatar System#

Generates 2D and 3D avatar representations from the persona's visual appearance references. Lip-sync animation derives viseme timing from the vocal stem's phoneme alignment; dance choreography animation maps beat, downbeat, and section boundaries (from @euterpe/core rhythm analysis) to motion keyframes. Hologram-ready export emits depth-layered geometry plus a synchronized audio track.

Social Media Management#

Comment and DM response generation operates within persona guardrails — replies must stay within the persona's personality traits and never assert real-world events. A fan-community model clusters fans by engagement tier and tenure to prioritize responses. Live-stream simulation produces a scripted but reactive-feeling stream: a setlist, between-song patter, and pre-generated responses keyed to common chat prompts. All outbound interactions are logged for provenance.

Virtual Performance#

Virtual concert generation assembles a setlist from the catalog (ordered by energy curve and key compatibility between adjacent tracks), a stage design, and crowd-interaction beats. The concert renders as a timed sequence of avatar performance, spatial audio (via @euterpe/spatial), and crowd simulation. Virtual meet-and-greet runs short scripted one-to-one interactions bounded by the same persona guardrails as social responses.

Business Management#

Aggregates streaming revenue per track and territory, tracks sync-licensing status (pitched, optioned, licensed, with fee and term), and matches brand partnerships by audience overlap and genre fit. NFT and digital-collectible creation, fan-token economy, and on-chain rights operations are delegated to @euterpe/chain. A&R pitch generation produces a one-page pitch from catalog performance and audience analytics for outbound label or sync opportunities.

Live Performance Engine (@euterpe/stage)#

Real-time performance tooling that operates under hard latency budgets. Unlike the offline @euterpe/studio graph, every stage operation is allotted a fixed per-buffer time slice and must degrade gracefully — dropping quality before dropping the deadline.

Ultra-Low-Latency Audio Pipeline#

A fixed-block processing graph targeting under 10 ms round-trip monitoring latency (input capture → processing → output) at a 64- or 128-sample buffer. The pipeline runs lock-free between the audio callback and control threads; any processor that overruns its slice is bypassed for that buffer and the overrun is logged. Latency compensation aligns all parallel processing chains to the longest chain so monitored audio stays phase-coherent.

Live Stem Separation#

Real-time separation into vocals, drums, bass, and other for live remixing, running a lightweight model variant of the @euterpe/genesis stem network. The renderQuality parameter trades separation cleanliness against latency (fast | balanced); high is reserved for offline work. Separation operates on a short rolling lookahead window, so the live stem feed lags the dry input by one window — disclosed to the performer so they can compensate.

Live Vocal Processing#

A real-time effects rack: pitch correction (transparent or hard "auto-tune" mode, with retune-speed and scale-snap parameters), harmonizer generating up to four diatonic harmony voices keyed to the live-detected key, and a configurable chain of compressor, reverb, delay, and de-esser. Shares processor implementations with @euterpe/voice vocal processing but runs them under the fixed-block budget.

Live Looping System#

A loop-station model: per-loop record, overdub, multiply, and clear, with loops quantized to a shared bar grid so layers stay locked. Loop states: empty → recording → playing → overdubbing → stopped. Overdub stacks new audio onto the existing loop without retriggering; an undo buffer reverts the last overdub layer.

Set and Setlist Management#

Orders performance items into a set with planned transitions between adjacent items — tempo bridge, key modulation route, or hard cut — derived from each item's tempo and key. Flags transitions that would jump more than a tritone in key or 20 BPM in tempo as abrupt so the performer can insert a bridge.

Live Notation Display#

Real-time scrolling notation synchronized to the transport, with an adjustable look-ahead so performers see upcoming bars. Auto-scroll follows tempo; a manual override holds position for rehearsal. Renders from @euterpe/core and @euterpe/theory notation data.

VR Concert Hosting#

Hosts virtual concerts in VR environments, combining @euterpe/spatial binaural/ambisonic rendering with avatar performance and audience presence. The audio scene updates with listener head position; the host scene state is the authority for source positions and acoustics.

AI Music Education (@euterpe/conservatory)#

Personalized learning for every instrument, voice, and theory concept — aligned to ABRSM, RCM, and other standard examination curricula.

Adaptive Learning System#

• Skill assessment engine: Entry-level skill testing to establish starting point.
• Learning path generation: Personalized curriculum based on current level and goals.
• Spaced repetition scheduling: Using memory science (from @mnemosyne/*) to optimize theory review.
• Practice time optimization: Recommending how to allocate limited practice time across skills.
• Parent/teacher dashboard: Progress visibility for oversight.
• Curriculum alignment: ABRSM grades 1–8 and diploma; Royal Conservatory of Music (RCM) levels.

Instrument Lessons#

• Piano: Keyboard visualization; MIDI input analysis with note accuracy, timing, and velocity feedback; fingering suggestions; sight-reading practice; classical, jazz, and contemporary repertoire.
• Guitar: Fretboard visualization; chord diagrams; audio input pitch accuracy analysis; strumming pattern detection; fingerpicking, barre chord, bending, vibrato, tapping, and slide technique; acoustic, electric, and bass curricula.
• Drums and percussion: Rhythm accuracy and timing precision analysis; rudiment library (the 40 standard snare drum rudiments); groove and fill libraries; independence exercises; foot technique.
• Voice: Real-time pitch visualization; vocal range assessment; breath control exercises; resonance training; register transition (chest/mix/head voice); vibrato development; style-specific technique.
• Additional instruments: Coverage for all major orchestral and band instruments.

Instrument lessons score a learner's MIDI or audio performance against the exercise reference on three axes: note accuracy (correct pitches played), timing (deviation from the grid in milliseconds), and dynamics (velocity match). An attempt carries these scores plus per-note error markers; the lesson advances when performance clears the level's pass threshold across consecutive attempts, preventing a single lucky run from unlocking progression.

Music Theory Curriculum#

Structured theory courses from note-reading basics through counterpoint, harmonic analysis, orchestration, and jazz theory. Each concept is a unit with an explanation, interactive exercises generated and graded by @euterpe/theory, and a unit quiz. Units have explicit prerequisites — harmonic analysis is gated behind interval and chord units — and a unit's state moves locked → available → in-progress → mastered only when the quiz is passed. Difficulty scales within a unit as the learner answers correctly.

Ear Training#

• Interval recognition: Identifying melodic and harmonic intervals by ear.
• Chord quality identification: Major, minor, diminished, augmented, dominant 7th, major 7th, minor 7th, half-diminished, fully diminished.
• Chord progression recognition: Identifying common progressions by ear.
• Melodic and harmonic dictation: Transcribing melodies and harmonies from audio.
• Rhythmic dictation: Transcribing rhythmic patterns from audio.
• Sight-singing: Singing unfamiliar notation on first reading.
• Critical listening for production: Frequency training (identifying EQ ranges); compression detection; critical listening skills for mixing and mastering work.

Exercises are generated by the @euterpe/theory ear-training module, which also tracks per-skill accuracy and adapts difficulty: a sustained accuracy above 85% widens the interval or chord vocabulary, below 60% narrows it. Sight-singing and dictation are graded by comparing the learner's pitch-tracked input to the target note sequence with a tolerance of ±50 cents per note.

Practice Tools#

A metronome with subdivision control (quarter, eighth, triplet, sixteenth clicks) and accented downbeats; a drone and chromatic tuner; a backing-track player; a chord-progression looper; a tempo trainer that raises tempo by a configurable step (default 5 BPM) each time a passage is played cleanly; a practice journal logging session duration and material; recording and playback; time-stretch slow-down that preserves pitch; and a scrolling notation display synced to the transport. The tempo trainer holds the new tempo only after a clean pass, reverting on a failed attempt so speed is earned, not assumed.

Film, TV, and Game Scoring (@euterpe/score)#

Composition tooling for picture and interactive media. Built from six sub-modules: video-analysis, film-scoring, adaptive-music, procedural-audio, and FMOD / Wwise game-audio export. The unit of work is a scored cue — a segment anchored to a timecode in/out range — and the score tracks leitmotif themes and their transformational variants.

Video Sync#

Spotting analyzes a video to propose cue points at scene cuts, dialogue boundaries, and emphasized actions, producing a candidate cue list the composer accepts or edits. SMPTE timecode sync locks the score timeline to picture at the project frame rate (24, 25, 29.97 drop/non-drop, 30 fps). The picture-lock workflow freezes the spotting timecodes; any post-lock picture edit raises a conform diff listing which cues shifted and by how many frames so the score can be re-conformed.

Adaptive Music for Games#

Exports cues as adaptive structures for FMOD and Wwise. Horizontal re-sequencing transitions between musical sections on game-state changes, quantizing the switch to the next bar or beat so transitions stay musical. Vertical remixing layers stems by intensity — each stem is assigned an intensity threshold and crossfades in or out as a game intensity parameter (0–1) crosses it. The export bundles per-section transition rules, stem layer thresholds, and sync markers in the target middleware's project format.

Temp Track Analysis#

Analyzes temp music placed by an editor to extract tonal and dramatic requirements: key, tempo, instrumentation, energy curve, and harmonic mood. The output is a brief that constrains original cue generation so the replacement score matches what the edit was cut to, without copying the temp track.

Leitmotif Management#

Maintains a registry of Leitmotif themes, each a short motivic cell with a character or concept association. The system tracks every transformational variant used across the score — transposition, mode change (major/minor), augmentation, diminution, fragmentation, reharmonization — using the @euterpe/theory motivic-analysis primitives, so a composer can audit where and how each theme appears and confirm dramatic payoffs land.

Orchestral Mockup to Live Session#

Converts a MIDI orchestral mockup into live-recording session preparation: per-instrument parts, transposed and within range (validated against @euterpe/theory orchestration ranges), divisi splits, click and pre-roll tracks, and a session order grouped by instrument family to minimize player idle time.

Style Analysis#

Analyzes the compositional style of reference film scores — harmonic language, orchestration density, thematic treatment — as a reference vocabulary for new work. It characterizes style for the composer's guidance and never reproduces the analyzed material.

AI Music Generation for Film and Games (@euterpe/ai-scoring)#

Multi-provider AI music generation and scoring orchestration for film and game production. Where @euterpe/score provides the composition tooling, ai-scoring is the layer that drives external AI music generators and shapes their output to picture and game state.

Provider Adapters#

Concrete adapters for the major AI music generators — Suno, Udio, AIVA, Boomy, and Google Lyria — each mapping Euterpe's neutral request types onto the provider's API: generation models, style and structure profiles, vocal modes, lyrics modes, license tiers, and job-state polling. A foundation module enumerates the available providers, capabilities, and workflows.

Generation Orchestration#

Multi-provider ensemble routing selects and combines generators per request; a generation cost optimizer weighs cost against quality; a prompt-to-score pipeline turns a creative brief into a generation plan. A scene-mood analyzer and tempo / intensity matching tie generation to picture and gameplay, and style-consistent generation keeps a cue set on-brand.

Scoring and Delivery#

A leitmotif system and an adaptive game-music system carry theme and intensity logic into interactive contexts; music-to-video sync aligns generated music to an edit; generated-music stem separation splits output for remixing; commercial-licensing verification checks usage rights; and MIDI export hands generated material to a DAW for refinement.

Real-Time Collaboration Platform (@euterpe/collab)#

@euterpe/collab provides multi-user editing primitives for a shared studio project. It handles synchronization, conflict-resolution, and version history; the host application supplies transport (WebSockets or similar) and durable persistence. Euterpe ships no server — the library gives you the data-structure and protocol layer, not the infrastructure.

Built from six sub-modules: audio-streaming, sync-engine, project-mgmt, rights-mgmt, video-conf, and file-sharing.

Low-Latency Audio Sharing#

Streams a performer's live audio to collaborators under a sub-100 ms one-way latency target — playable for loose tracking though not for tight ensemble. Audio is sent at a low-latency codec setting; a jitter buffer absorbs network variance and reports its current depth so a session can decide whether live tracking is viable or asynchronous overdubbing is the better mode.

Synchronized DAW Sessions#

Multiple users edit the same project simultaneously. Project state is modeled as a CRDT (conflict-free replicated data type) so concurrent edits from different clients converge to the same result without a central lock. Each user holds a soft selection (the clips or parameters they are editing); other users see those as live presence highlights to reduce accidental collisions.

Conflict Resolution#

Edits are commutative deltas (move clip, change parameter, add track). Two edits to disjoint objects always merge cleanly. Two edits to the same scalar parameter resolve last-writer-wins by logical timestamp; two structural edits to the same clip (e.g., both trim it) merge by composing the deltas where possible and otherwise flag a conflict for one user to resolve. The resolution outcome is deterministic given the same delta set on every client.

Version History#

Every applied delta is appended to an ordered history. Any user can roll the project back to a prior point; rollback is itself recorded as a delta so it replicates to all clients and is undoable. Named checkpoints mark milestones (e.g., "vocal comp done") for fast navigation.

Video Call Integration#

Carries a synchronized video and voice channel alongside the session for visual communication. The video channel is independent of audio sharing — it never competes with music audio for the latency budget.

Asynchronous Collaboration#

For collaborators in different time zones, supports leaving stems, alternate takes, and timestamped notes anchored to positions on the timeline. A collaborator opening the project later sees pending contributions as a review queue and accepts or rejects each into the shared project.

Music Intelligence and Recommendation (@euterpe/discover)#

Music Information Retrieval (MIR) and recommendation logic. Built from five sub-modules: audio-features, semantic-analysis, similarity, recommendation, and playlist. It works from per-track extracted features and a semantic embedding, reused across all discovery operations.

Acoustic Fingerprinting#

Generates a compact acoustic fingerprint that identifies an audio clip regardless of format, bitrate, or encoding, and matches it against a catalog of known fingerprints. Identification is robust to partial clips and moderate time/pitch shifts. This identifies tracks; copyright adjudication is out of scope and delegated to Themis.

Feature Extraction#

Extracts the standard MIR feature set per track: tempo (BPM), key, energy, danceability, valence (musical positivity), acousticness, instrumentalness, and loudness. Each scalar feature is normalized 0–1 except tempo and key. Features are deterministic for a given input so the same track always yields the same feature set.

Semantic Similarity#

Represents each track as a high-dimensional embedding capturing musical similarity beyond surface acoustics — two tracks can be acoustically different yet semantically close (same mood, same scene-fit). Similarity is cosine distance between embeddings; nearest-neighbor search returns ranked similar tracks.

Context-Aware Recommendations#

Recommends what to play next from listener context — stated mood, time of day, and activity — biasing feature targets accordingly (e.g., higher energy for workout context, lower for focus). Recommendations combine the context target with the current track's profile so the next pick is both situationally apt and a smooth follow-on.

Discovery Algorithms#

Balances familiarity and novelty with an explicit exploration parameter (0 = only close matches, 1 = maximally novel within the listener's tolerance). The default sits mid-range so a session mixes recognizable picks with discovery without jarring the listener.

Playlist Generation#

Builds a coherent playlist with smooth transitions: tracks are ordered so adjacent pairs have compatible key (within the circle of fifths) and a gradual rather than abrupt energy change, producing an arc rather than a shuffle. The caller sets length and an optional energy curve the ordering follows.

Blockchain and Web3 Music Infrastructure (@euterpe/chain)#

@euterpe/chain owns on-chain rights and royalty primitives. It composes transactions and verifies on-chain state; signing and broadcast are the host application's responsibility, and Euterpe holds no private keys or wallets. The library boundary is intentional: Euterpe owns the music-rights data model; the Aje domain owns cryptocurrency and Web3 infrastructure.

Royalty Tracking#

Records the royalty split for a work as on-chain shares — each rights holder's percentage, with all shares summing to 100% — so every party can independently verify how revenue is calculated and distributed. The split is versioned; changing it requires a new version that does not retroactively alter payouts already settled under the old one.

Smart Contracts for Streaming#

Generates and deploys royalty-distribution contracts that pay rights holders automatically when play events are reported on-chain. Each play event credits the splits defined in the current royalty version. Payments below a configurable dust threshold accumulate rather than triggering an uneconomic micro-transaction per play.

Fractional Music Ownership#

Tokenizes a song's royalty rights into fungible shares so fans can own a percentage of future royalty income. A holder's payout is proportional to their share of the float. Ownership transfers update the entitled-payee set automatically for subsequent distributions.

NFT Minting#

Mints unique digital ownership certificates for master recordings, unreleased tracks, individual stems, and live recordings. Each NFT carries metadata linking to the underlying asset and its @euterpe/provenance record so authenticity is verifiable. Minting records the creator and an edition size (1 for a unique, N for a limited run).

Fan Token Creation#

Issues an artist-specific fungible token for fan engagement and governance — usable for gated access, voting on artist decisions, and rewards. Supply, distribution schedule, and governance weighting are configured at creation.

Secondary Market Infrastructure#

Enables verified resale of music NFTs with provenance carried across each transfer, so a buyer can trace the full ownership chain back to the original mint. Resale can route a configurable royalty percentage back to the original creator on every secondary sale.

Rights Registry#

Maintains an immutable on-chain record of composition and master ownership — the authoritative claim of who owns what. Registry entries are append-only; corrections are new entries that supersede prior ones while leaving the full history auditable.

AI Sound Design and Synthesis (@euterpe/synth)#

Sound-design generation and analysis that targets the synthesizer instruments defined in @euterpe/studio (subtractive, FM, wavetable, granular), producing patches and wavetables those engines can load directly.

Text-to-Sound-Effect#

Generates non-musical sound effects from a text description — impacts, risers, ambiences, foley — with a requested duration. Output is a rendered audio buffer plus a one-shot/loopable classification so the effect can drop straight into a sampler or timeline.

Synthesizer Patch Generation#

Produces a synthesizer patch from a text description ("warm analog pad", "aggressive FM bass") as a structured parameter set for the target engine — oscillator, filter, envelope, and modulation values — rather than baked audio, so the result stays fully editable. The target engine is specified; the generated parameters are validated against that engine's ranges.

Timbre Analysis#

Decomposes any sound into timbral constituents — spectral centroid, harmonic vs. inharmonic content, attack sharpness, and a descriptive tag set (bright, hollow, metallic, breathy). The analysis drives patch generation and cross-synthesis and gives a measurable target for matching a reference sound.

Cross-Synthesis#

Combines the timbral qualities of two sounds — applying the spectral envelope of one to the harmonic structure of another, or morphing continuously between them with a 0–1 blend parameter — to create hybrid timbres.

Wavetable Design#

Creates custom wavetables for wavetable synthesis: an ordered series of single-cycle waveforms designed so scanning through the table produces a smooth spectral evolution. Wavetables export in the standard frame layout the @euterpe/studio wavetable instrument expects.

Physical Modeling#

Simulates the acoustics of physical instruments — string, membrane, air-column, and bar models — parameterized by physical properties (string tension and length, body resonance, excitation type). Physical modeling yields natural playing dynamics that sample playback cannot, since timbre changes continuously with the simulated excitation.

AI Songwriting Assistant (@euterpe/lyrics)#

Lyric authoring that treats words and music as one problem — every lyric operation can be constrained by a melody so syllables, stresses, and phrase lengths align with the music rather than fighting it.

Lyric Generation From Prompts#

Generates a full lyric from a theme, mood, or story prompt, structured to a requested form (verse-chorus, AABA, through-written) with each section's role respected — verses advance narrative, the chorus restates the central idea. When a melody is supplied, generation is constrained to its phrase lengths.

Rhyme Scheme Assistance#

Suggests rhymes that are both phonetically valid and semantically on-topic, so suggestions advance the lyric's meaning rather than forcing a non-sequitur for the sound. Supports perfect, slant, and internal rhyme, and tracks the active rhyme scheme (ABAB, AABB, etc.) so suggestions fit the established pattern.

Syllable Counting#

Counts syllables per line and aligns them to melodic phrase lengths: given a melody, it reports whether each line has too few or too many syllables for its phrase and proposes contractions or word swaps to fit. The target is one note per syllable unless melismas are explicitly allowed.

Meter Analysis#

Analyzes the stress pattern of a lyric line against the melodic rhythm, detecting where a stressed syllable lands on a weak beat or an unstressed syllable on a strong beat — the mismatches that make a lyric feel awkward to sing — and flags them for revision.

Genre-Specific Vocabulary#

Biases word choice, imagery, and phrasing toward a target genre's conventions so a lyric reads as authentic to that style rather than generic.

Hook Generation#

Specializes in the hook — the single most memorable line. Generates hook candidates optimized for brevity, repeatability, and a strong stressed-syllable landing on the chorus downbeat, and ranks them by memorability.

Multi-Language Lyrics#

Writes lyrics in any supported language, respecting that language's natural stress and syllable structure rather than translating an English meter onto it. Output flagged for cultural-authenticity review where idiom or reference may not carry across languages.

Co-Writing Mode#

Operates as a collaborator rather than a replacement: the human writes lines and the assistant offers alternatives, completions, rhyme options, and meter fixes for the human to accept, reject, or edit. Authorship of each line is tracked so the human stays in control of the final lyric.

Copyright Protection and Content Detection (@euterpe/protect)#

Detection analysis for music provenance and originality. Every check returns a structured verdict — a score, a confidence, and the features that drove the result — never a bare boolean, so downstream callers can set their own thresholds and surface explanations in disputes. @euterpe/protect analyzes and scores; legal copyright adjudication is delegated to Themis (see Domain Boundaries). It also certifies streamer-safe replacement music and originality for the V2 karaoke bridge (karaoke-rhythm-combat-bridge; see specifications.md §12, Cross-Domain Integration).

Audio Fingerprinting#

• Chromaprint fingerprinting: Generating compact acoustic fingerprints that identify any audio regardless of format, quality, or encoding.
• Fingerprint matching: Matching against a catalog of registered works, including time-shifted, pitch-shifted, and speed-changed matching that defeats simple fingerprint evasion.
• Cover song detection: Identifying when a recording is a cover of a registered work even when the performance — instrumentation, tempo, key — is substantially different, by matching melodic and harmonic structure rather than the surface waveform.
• Real-time fingerprint matching: Live identification during streaming or upload, matching on a rolling window so a match is reported within seconds of the identifying section playing.

A fingerprint match returns the matched work, the matched time range in both recordings, and a confidence; the caller decides whether the confidence clears its action threshold.

AI-Generated Music Detection#

• AI vs. human classifier: Classifying whether music was AI-generated, distinguishing Suno, Udio, AIVA, and similar generator outputs from human-performed recordings.
• Spectral artifact detection: Finding the characteristic spectral artifacts that AI music generators leave.
• Micro-timing analysis: Human performers have subtle micro-timing irregularities that AI generation typically lacks or replicates unnaturally; the distribution of timing deviations is a strong signal.
• Explainability: Naming which features drove the AI verdict — required for dispute resolution, since an unexplained verdict cannot be contested.

The verdict is a likelihood (0 = confidently human, 1 = confidently AI) with the contributing features ranked by weight, not a hard label, so a borderline result is visibly borderline.

Voice Deepfake Detection#

• Voice authenticity verification: Classifying whether a vocal performance is a genuine human recording or synthesized.
• Synthesis artifact detection: Finding artifacts left by voice-cloning tools (RVC, So-VITS-SVC, ElevenLabs).
• Breath pattern analysis: Natural breathing — its timing, depth, and placement between phrases — is among the hardest things for voice cloning to replicate, and an absent or unnatural breath pattern is a strong indicator.
• Consent verification: Cross-referencing whether a ConsentRecord exists for the voice model of a real artist — connecting detection to the consent-gating that @euterpe/voice enforces structurally.

A vocal flagged as synthetic with no matching consent record is the actionable case; the verdict carries the synthesis likelihood and the consent-record lookup result so the two signals are visible separately.

Plagiarism and Similarity Detection#

• Melodic similarity detection: Scoring melodic overlap between a new work and a catalog, robust to transposition and tempo change so a copied melody in a different key is still caught.
• Harmonic progression matching: Detecting shared chord progressions, weighting distinctive progressions over ubiquitous ones (a I-V-vi-IV match carries far less weight than an unusual chromatic sequence).
• Pre-release clearance: Checking a finished work against a catalog before release and returning the closest matches with their similarity scores, so a risky resemblance is caught before publication.
• Legal precedent database: Contextualizing a similarity score against documented infringement cases to estimate risk band — informational only; the legal determination is delegated to Themis.

Similarity output is a ranked match list with per-match melodic and harmonic scores and the overlapping passage, never a single infringe/clear flag.

Music Video and Audio-Reactive Visuals (@euterpe/video)#

Visual generation driven by musical structure. Every operation consumes the beat, measure, and section analysis from @euterpe/core and @euterpe/theory, so visuals are anchored to the music's actual structure rather than a fixed clock.

Beat-Synced Video Generation#

Generates AI video where cuts, transitions, and motion land on musical boundaries — cuts on downbeats or section changes, motion accents on beats. The edit decision list is built from the analyzed grid first, then video is generated to fill each segment, guaranteeing sync regardless of the generator's output timing.

AI Video Generation#

Produces a full music video from the song and a style description, segmented by song section (intro, verse, chorus, bridge) so each section gets distinct visual treatment while a carried style reference keeps the video coherent across cuts.

Lyric Video Generation#

Generates typography-animated lyric videos with kinetic text and karaoke-style word highlighting timed to the vocal. Word timing comes from the vocal stem's phoneme alignment. Output renders in vertical (TikTok, 9:16), horizontal (YouTube, 16:9), and square (Instagram, 1:1) variants, each with layout tuned to its aspect ratio rather than a cropped master.

Audio-Reactive Visuals#

Real-time visuals responding to frequency content, amplitude envelope, and beat detection — waveform animations, spectrum analyzers, particle systems, and 3D scenes. Each visual exposes its mapping (which audio feature drives which visual parameter, and the response curve) so reactions can be tuned per track.

VJ Software Integration#

Exports scenes and reactive patches to Resolume, TouchDesigner, and OBS in each tool's project format, so a generated visual set drops into an existing live VJ or streaming rig.

Sample Library and AI Loop Generation (@euterpe/samples)#

Sample and loop generation, organization, and search. Built from four sub-modules: sample-gen, beat-gen, sample-search, and sample-market. Samples carry musical metadata — type, tempo, key (where pitched), and timbral tags — so they can be matched to a session and found later.

AI Sample Generation#

Text-to-sample generation for drum one-shots, melodic loops, bass loops, chord loops, and FX (risers, impacts, sweeps). The caller specifies type and length; loop-type output is generated seamlessly so the end joins the start without a click, and one-shot output is generated with a clean decay to silence.

Groove-Matched Loops#

Generates loops that match the tempo and groove of an existing session — the loop locks to the session's timing feel (swing ratio, micro-timing) extracted by @euterpe/theory groove analysis, so it sits in the pocket rather than fighting the existing rhythm.

Key-Matched Loops#

Generates pitched loops in the current project's key, or transposes an existing loop into that key, so dropped-in melodic and chord loops are harmonically consistent without manual pitch correction.

Sample Marketplace#

Provides the model for community sharing and discovery of generated samples — each sample's musical metadata plus provenance — so listings are searchable by musical attributes. Hosting and transactions belong to the consuming application; Euterpe ships no marketplace server.

Stem Extraction From Samples#

Separates an existing sample or loop into individual instrument stems via the @euterpe/genesis stem network, so a found loop can be split and its parts reused independently.

Sample Pack Organization#

Organizes samples into packs with rich metadata — type, tempo, key, tags, provenance — and indexes them for fast attribute search ("120 BPM A-minor melodic loops"), keeping a large library findable.

Audio Restoration and Remastering (@euterpe/restore)#

Restoration processors for damaged and archival audio. Each processor exposes a strength control and runs non-destructively against a source copy, since over-restoration is itself a defect — removing real signal along with the artifact.

Noise Reduction#

Removes broadband noise floor (tape hiss, room noise) using a learned model of noise versus signal, attenuating noise while leaving musical content intact. A noise profile can be captured from a silent passage to target the specific noise in the recording.

Click and Crackle Removal#

Detects and repairs the transient impulse defects of vinyl and tape — clicks, pops, crackle — by interpolating the damaged samples from surrounding audio. Sensitivity is tunable so genuine percussive transients are not mistaken for clicks.

Hum Removal#

Removes mains hum at 50 Hz or 60 Hz and its harmonic series using narrow notch filters tuned to the fundamental and overtones, eliminating the hum with minimal effect on nearby musical content.

De-Reverb#

Reduces unwanted room reverberation from recordings made in poor acoustic spaces, separating the direct signal from the reverberant tail. Strength is adjustable because fully dry audio sounds unnatural; the goal is a controlled reduction.

Bandwidth Extension#

Synthesizes plausible high-frequency content for recordings that lack it (old formats, low-bitrate sources), restoring perceived air and clarity. The synthesized band is generated from the harmonic structure of the existing audio so it is consistent with the source rather than generic.

Stereo Reconstruction#

Creates a natural-sounding pseudo-stereo image from a mono source — adding width while preserving mono compatibility so the result does not collapse or phase-cancel when summed.

Historical Recording Remastering#

Chains the above into a complete remastering pipeline for archival recordings — denoise, de-click, de-hum, bandwidth extension, then a gentle final balance — with each stage's strength tuned to the recording's condition and the pipeline order fixed so corrections do not interfere.

Music Analytics and Business Intelligence (@euterpe/analytics)#

Music business intelligence over performance data ingested from DSPs and distributors. The library computes metrics and models; it does not store data — the host application owns ingestion and persistence.

Streaming Analytics#

Aggregates per-track and per-release performance across all DSPs into one normalized view — streams, listeners, saves, skips, completion rate, and playlist adds — reconciling each platform's differing metric definitions so cross-DSP totals are comparable.

Audience Demographics#

Breaks the listener base down by age band, gender, geography, and listening behavior (repeat rate, session context, discovery source), so an artist can see who is listening and how.

Revenue Attribution#

Resolves revenue to its source — which DSP, which territory, which time period, and which release — and reconciles reported payments against expected payouts from stream counts, surfacing underpayment discrepancies for follow-up.

Trend Detection#

Identifies momentum early by tracking growth velocity and acceleration rather than absolute totals, so a track gaining fast from a small base is flagged before it reaches mainstream scale.

Comparison Benchmarking#

Compares a release against a peer cohort (similar genre, audience size, and career stage) so performance is read relative to a fair baseline rather than in isolation.

Prediction Modeling#

Forecasts streaming performance from early signals — first-week velocity, save rate, playlist placement, skip rate — producing a projected trajectory with a confidence band so the forecast's uncertainty is explicit.

Audio-to-Notation Transcription (@euterpe/transcribe)#

Audio-to-symbolic conversion — turning a recording into editable notation, MIDI, or chord charts. Transcription is inherently uncertain; every output carries a per-note confidence so a user can see and correct low-confidence regions rather than trust a flat result.

Polyphonic Transcription#

Transcribes multi-instrument audio into notation — detecting concurrent pitches, onsets, durations, and instrument assignment. As the hardest problem in music information retrieval, results improve with cleaner source material; a stem separation pass (via @euterpe/genesis) before transcription raises accuracy on dense mixes.

Drum Transcription#

Converts a drum recording into drum notation, classifying each onset by drum voice (kick, snare, hi-hat open/closed, toms, cymbals) and placing it on the rhythmic grid.

Guitar Tablature Generation#

Produces playable guitar tablature from guitar audio — not just pitches but string-and-fret assignments chosen for a physically sensible fingering, so the tab is actually playable rather than a literal pitch dump.

Chord Chart Generation#

Analyzes harmony from audio and produces a chord chart — chord symbols aligned to bars — using the @euterpe/theory harmonic-analysis primitives.

MIDI Output#

Exports the transcription as MIDI for further editing — pitches, onsets, durations, and velocities — so a transcribed part can be opened directly in @euterpe/studio or any DAW.

MusicXML Output#

Exports MusicXML for import into Sibelius, Finale, and MuseScore, preserving notation detail (clefs, key and time signatures, beaming, ties) beyond what MIDI carries.

Sacred Music and Sound Healing (@euterpe/sacred)#

A knowledge base of world sacred-music traditions plus a generator for sound-healing audio. Generated healing audio is presented as a wellness tool; the library makes no medical claims and labels every preset accordingly.

World Sacred Music Traditions#

A structured catalog of sacred-music traditions — Gregorian chant, Byzantine chant, Carnatic and Hindustani classical music, Sufi music (Qawwali, sama), Jewish liturgical music (cantorial tradition, Ashkenazi and Sephardic nusach), Gospel, and Buddhist chant (Tibetan overtone singing, Theravada chanting). Each tradition entry records its characteristic modal system (mapped onto the @euterpe/core scale catalog where one exists — raga, maqam, the church modes), its rhythmic conventions, performance context, and notation system, so a tradition can both inform analysis and constrain generation.

Sound Healing Modalities#

Generators for established sound-healing forms: binaural beats (two tones a small frequency difference apart, one per ear, producing a perceived beat at the difference — delta 0.5–4 Hz, theta 4–8 Hz, alpha 8–13 Hz, beta 13–30 Hz); isochronic tones (a single tone pulsed at a target rate); Solfège-frequency drones (396, 417, 528, 639, 741, 852 Hz); singing-bowl synthesis with rich inharmonic partials and long decay; and cymatics-informed composition built on simple frequency ratios. Each generator takes a target rate or frequency and a duration and renders directly to an audio buffer.

Therapeutic Applications#

Curated presets for meditation, sleep, focus, stress reduction, and pain management, each combining a modality, a frequency or beat rate, and a tempo, with the rationale cited. Presets are advisory wellness content, not a medical prescription.

Music History Database (@euterpe/history)#

A structured, queryable knowledge base of music history and ethnomusicology. Entities — periods, genres, traditions, composers, instruments — are linked into a graph so the database answers relational questions (influence, descent, contemporaneity), not just lookups.

Comprehensive Timeline#

A timeline from ancient music (Mesopotamian hymns, Greek modes) through the medieval, Renaissance, Baroque, Classical, Romantic, and 20th-century periods to contemporary music. Each period entry records its date range, defining stylistic traits, representative forms, and characteristic harmonic and rhythmic practice, so any work or composer can be situated in context.

Genre Genealogies#

A descent graph of how genres evolved from and influenced one another — the jazz → R&B → rock → hip-hop lineage, the folk → country → rockabilly → rock and roll progression. Each edge records the direction and nature of influence, so the graph can be traversed to trace a modern genre back to its roots.

Ethnomusicology Database#

Documents music traditions of cultures worldwide — instruments, scales, performance contexts, and social functions — with each tradition's scale system cross-linked to the @euterpe/core scale catalog so its modal material is usable, not just described.

Composer and Artist Database#

Biographies, discographies, and influence networks for composers and artists. The influence network is part of the same graph as genre genealogy, so an artist's stylistic lineage and a genre's lineage connect.

Instrument History#

Records how instruments were invented, evolved, and spread across cultures — construction, playing technique, tuning, and geographic and temporal spread — linked to the traditions and periods in which each instrument was central.

Music Philosophy and Aesthetics (@euterpe/philosophy)#

A structured reference base on the philosophy, aesthetics, and ethics of music. It organizes positions, arguments, and thinkers into linked entries so a position can be traced to its proponents and counter-positions — a navigable knowledge base, not a computational engine.

Philosophy of Music#

Catalogs the central debates: formalism (Hanslick — music expresses nothing beyond itself) versus expressivism (music expresses emotion); the ontology of the musical work (what a piece of music is — score, performance, or abstraction); and theories of musical meaning and reference. Each position entry links to its proponents, its standard arguments, and the positions it opposes.

Aesthetics of Music#

Documents theories of musical beauty, the musical sublime, and the music-emotion relationship — how and whether music conveys feeling. Entries are cross-linked to the philosophy-of-music positions they presuppose.

Ethics of Music#

Covers AI authorship (who, if anyone, authors AI-generated music), cultural ownership, and the appreciation-versus-appropriation distinction — the ethical questions Euterpe's own generation libraries raise. Entries connect to the @euterpe/provenance and @euterpe/protect concerns they bear on, so the ethical framing sits alongside the technical controls.

Music and Identity#

Documents how music constructs ethnic, national, gender, and subcultural identity — the social function of musical style and genre as identity markers.

Acoustic Science and Psychoacoustics (@euterpe/acoustics)#

Acoustic and psychoacoustic computation — the physics and perception layer underneath the spatial, mastering, and accessibility libraries. Every function is a deterministic calculation over physical inputs, returning measured values rather than estimates.

Room Acoustics#

Computes room behavior from geometry: axial, tangential, and oblique room modes (standing-wave resonant frequencies determined by room dimensions), RT60 reverberation time from volume and absorption, and reflection, diffusion, and absorption coefficients. Acoustic-treatment design recommends absorber and diffuser placement to flatten problem modes toward a target RT60. The room models feed @euterpe/spatial VR/AR reverberation.

Psychoacoustic Effects#

Models how hearing processes sound: the Fletcher-Munson equal-loudness contours (loudness perception varies with frequency and level), frequency masking (a loud tone raises the audibility threshold of nearby quieter tones), the precedence effect (localization is dominated by the first-arriving wavefront), and auditory streaming (the perceptual separation of simultaneous sources). These models inform mix and EQ decisions in @euterpe/master and @euterpe/studio.

Loudness Models#

Computes integrated, short-term, and momentary LUFS to the ITU-R BS.1770 standard, true-peak level, dynamic-range metrics, and the gap between perceived loudness and measured SPL. Provides the measurement layer the mastering library's loudness-normalization targets are checked against.

Hearing Protection#

Computes safe exposure time at a given SPL on the standard 3 dB exchange rate (a 3 dB increase halves the safe duration) and estimates cumulative hearing-damage risk from a session's level-over-time profile — supporting the haptic and volume-aware accessibility features.

Provider and Infrastructure Layer#

Provider Connectivity (@euterpe/providers) (planned)#

The model-selection layer that @euterpe/genesis calls to pick and reach an AI music model. It abstracts every backend behind one interface so generation code never names a specific provider.

• Model capability registry: A registry entry per available model (MusicGen, Stable Audio, Lyria, Suno, etc.) recording the tasks it supports (text-to-music, stems, voice, inpainting), its constraints (max duration, supported sample rates), and its cost, typical latency, and quality tier. Entries are updated as models change; an entry can be marked unavailable to remove a model from routing without code changes.
• Intelligent routing: For each request, filters the registry to models that support the task and constraints, then ranks the candidates against the request's quality requirement, latency budget, and cost ceiling, selecting the best feasible model. A request that no model can satisfy fails fast with the unmet constraint named rather than silently downgrading.
• Fallback chains: Each route has an ordered fallback list. If the primary model is unavailable or returns output below the quality threshold (scored via @euterpe/evals), routing advances to the next candidate. State per attempt: selected → calling → succeeded | failed → fallback. The chain is exhausted only when every candidate fails, and the failure reports every model tried.
• Cost optimization: When a request's quality requirement is loose (a draft or sketch), routing prefers a fast, inexpensive model; a final-master request routes to a premium model. Per-request cost is recorded for @euterpe/ops cost governance.
• Streaming support: For models that generate progressively, relays audio chunks to the caller as produced rather than buffering to completion, so preview playback can start before generation finishes.

ElevenLabs Integration (@euterpe/elevenlabs) (planned)#

A concrete provider adapter for ElevenLabs' music and voice generation, registered with @euterpe/providers so the router can select it. It maps Euterpe's neutral request types onto the ElevenLabs API and back: music generation with full parameter pass-through, voice-library management (browsing and selecting stock voices), streaming generation relayed chunk-by-chunk, and the Voice Design API for creating original synthetic voices from a description. A voice created here is registered with a ConsentRecord (the @euterpe/voice consent type), so it flows through the consent-gated voice operations like any other.

Google Lyria Integration (@euterpe/lyria) (planned)#

A concrete provider adapter for Google's Lyria music model, reachable via either OpenRouter or the direct Google AI API, registered with @euterpe/providers. Supports all Lyria generation parameters and model variants, and exposes Lyria's distinguishing real-time interactive generation — a continuous session whose output steers as prompt parameters change mid-stream, surfaced through the same streaming interface as other progressive providers. The transport (OpenRouter vs. direct) is selectable per deployment and transparent to callers.

Project Persistence (@euterpe/projects) (planned)#

The data model for an Euterpe project and its asset history. It defines the project structure and version semantics; durable storage is the host application's responsibility (Euterpe persists nothing — see Domain Boundaries).

• Asset graph: A directed acyclic graph whose nodes are project assets (audio files, MIDI sequences, lyrics, images, stems) and whose edges are derivation relationships — which stems came from which generation prompt, which mix from which stems, which master from which mix. Because the graph is acyclic and edges record the producing operation, any asset's full lineage is traversable, giving a complete audit trail. The graph feeds @euterpe/provenance.
• Version control: Every project mutation produces a new immutable version; the version history is an append-only chain supporting rollback to any prior state and named checkpoints. Rollback creates a new version rather than deleting history, so editing is non-destructive at the project level and a rollback is itself reversible.
• Collaboration state: Models multi-user project access with optimistic concurrency — each client edits a version and submits; a submit against a superseded version is rejected with a conflict so the client can rebase. This is the persistence-side counterpart to the real-time CRDT model in @euterpe/collab.

Generation Workflows (@euterpe/workflows) (planned)#

The orchestration runtime that chains individual generation and editing steps into a reliable multi-step pipeline.

• Generation orchestration: Runs multi-step generation pipelines — for example text → stems → mix → master — as a directed sequence of steps with declared inputs and outputs. Each step's result is checkpointed; on interruption or partial failure the workflow resumes from the last completed checkpoint rather than restarting, so a late-stage failure never discards early-stage work. Step states: pending → running → completed | failed; a failed step is retried per its policy before the workflow halts. Every step output is registered into the @euterpe/projects asset graph.
• Editing runtime: A stateful environment for non-destructive editing in which every edit is a reversible delta against the source rather than a mutation of it. The delta log gives unbounded undo/redo and lets a sequence of edits be replayed, reordered, or partially reverted. Source audio is never overwritten — the rendered result is always the source plus the current delta set.

Provenance and Rights (@euterpe/provenance) (planned)#

The rights and provenance ledger for project assets. It records how each asset came to be and what may legally be done with it, and enforces those constraints at generation time.

• Generation provenance tracking: For every generated asset, records the exact model, prompt, negative prompt, seed, and parameter settings that produced it, plus the source assets it derived from. The record is written at generation time and is immutable, so an asset's origin is fully reconstructible and the same record makes a generation reproducible — re-running it with the recorded inputs yields the same output.
• Rights management: Tracks ownership and licensing per asset, classifying each as wholly AI-generated, AI-assisted human work, or human-authored, and attaching any license terms and restrictions. Restrictions propagate along the asset graph: a derivative inherits the most restrictive license among its inputs.
• Compliance runtime: Enforces rights constraints before a generation commits — blocking a restrictively-licensed sample from a commercial project, and flagging when generated output scores too similar to a registered work (similarity scoring delegated to @euterpe/protect; copyright adjudication to Themis). A violation stops the operation and names the offending asset and rule.

Quality Evaluation (@euterpe/evals) (planned)#

The quality-measurement layer for generated audio. It produces the scores that @euterpe/providers routing and the release gates depend on. The @euterpe/genesis quality sub-module already provides per-track quality, musicality, coherence, and artifact-detection scoring; @euterpe/evals is the planned standardized-benchmark and release-gate layer on top of it.

• Quality benchmarks: Standardized metrics per generation task — Fréchet Audio Distance (FAD) and KL divergence against a curated gold-standard set, plus a listener-preference model — combined into overall quality, musical coherence, artifact-level (lower is better), and perceptual-quality scores. The same benchmark set is reused across runs so scores are comparable over time.
• Regression testing: A fixed evaluation suite re-run after every model update against the prior version's scores. A drop beyond the configured regression tolerance on any tracked metric fails the suite and blocks the update from reaching production, so a model upgrade cannot silently degrade output.
• Release gates: A configurable minimum-score gate that a track must clear before release through @euterpe/distribution. A track scoring below the floor is held with the failing metrics named; the gate enforces a uniform quality floor across all Euterpe outputs.

Observability and Operations (@euterpe/ops) (planned)#

The operational instrumentation layer for the generation libraries — the telemetry, cost-control, and reliability primitives a host application wires around AI model calls.

• Observability: Distributed tracing, metrics, and structured logging across generation operations. A trace spans the full pipeline (route → model call → post-process → evaluate), so latency can be attributed to a specific stage and a failure tied to the step that caused it.
• Cost governance: Attributes each model call's cost (reported by @euterpe/providers) to a project and user and aggregates running spend. Configurable budget thresholds raise an alert as spend approaches a cap and a hard stop at the cap, preventing an automated pipeline from running away.
• Reliability: Circuit breakers, retry policies, and health monitoring for external model integrations. A circuit breaker trips a model out of rotation after a configured failure rate and lets @euterpe/providers route to fallbacks; it half-opens to probe recovery before restoring the model. Retry policies use bounded exponential backoff so a struggling provider is not hammered.

API Service Layer (@euterpe/api) (planned)#

Service-interface primitives — typed contracts and middleware — that a host application mounts to expose Euterpe over the network. Euterpe itself ships no running server, database, or event bus (see Domain Boundaries); this library provides the contracts and request-handling logic, and the host supplies the HTTP/WebSocket runtime.

• Service contracts: Typed contracts for every Euterpe service operation — input schema, response shape, and an enumerated error set — with explicit versioning so a contract change is a deliberate, visible event rather than a silent break for clients.
• Realtime delivery: WebSocket and Server-Sent Events handlers for streaming generation results, relaying audio chunks to the client as a progressive provider produces them so playback can begin before generation completes.
• Rate limiting and quotas: Per-user and per-project request-rate limits and generation quotas, rejecting calls over the limit so an automated pipeline cannot run away on cost or starve other users.
• Authentication middleware: JWT and API-key authentication applied uniformly across operations, so every service call is authenticated by the same path.

Studio Runtime (@euterpe/studio-runtime) (planned)#

The opinionated product layer above the raw generation primitives. Where @euterpe/genesis and @euterpe/master expose granular operations, this library packages them into high-level, intent-driven operations a product can call directly.

• Directed SOTA workflows: End-to-end workflows that chain multiple models into one operation — "generate a complete song from a text description" internally runs text-to-music, stem separation, mixing, and mastering as a single @euterpe/workflows pipeline with the model selection, ordering, and quality gates already decided.
• Product experience layer: Translates a high-level user intent into the concrete sequence of model calls, post-processing steps, and quality gates needed to satisfy it — so the product expresses what the user wants and the runtime owns how it is produced.
• Workflow templates: Reusable, parameterized workflow definitions for common production tasks — single production, EP production, soundtrack batch generation — each a named template a product instantiates with its specific inputs rather than re-assembling the step graph.

Multi-Platform Release Management (@euterpe/distribution)#

Music distribution across every major DSP (Digital Service Provider — music streaming platforms) and physical channel. Built from four sub-modules: dsp-integration, distributor, metadata, and analytics-reporting. The unit of work is a release (release types single / ep / album / compilation / deluxe / live). Metadata and asset checks gate delivery; a per-DSP delivery that fails is flagged with its rejection reason so it can be corrected and resubmitted without affecting platforms that succeeded.

• DSP integrations: API integrations with Spotify for Artists, Apple Music for Artists, Amazon Music, YouTube Music, Tidal, Deezer, SoundCloud, Bandcamp, TikTok/Resso, Instagram Music, and regional platforms (NetEase Cloud Music, QQ Music) — one submission fans out to every selected market, each platform receiving assets in its required format.
• Distributor APIs: Integration with aggregators (DistroKid, TuneCore, Amuse, RouteNote) for artists who prefer a single aggregator over direct DSP relationships — the same release routes through an aggregator instead of fanning out to DSPs directly.
• Metadata management: ISRC (International Standard Recording Code) generation per track, ISWC (International Standard Musical Work Code) for compositions, and UPC management per release, validated against each DSP's metadata specification before delivery so a release is not rejected downstream for a metadata error.
• Release scheduling: Embargo scheduling for a coordinated global release at an exact instant, with timezone-aware delivery timed so every DSP goes live simultaneously rather than rolling out across the day.
• Content ID registration: Registering audio fingerprints with YouTube Content ID, Facebook Rights Manager, and other UGC rights-management systems, enabling automatic monetization or takedown of infringing user uploads.
• Podcast and audiobook distribution: Extends the same release model to spoken-word platforms — Spotify Podcasts, Apple Podcasts, Audible.
• Physical distribution coordination: Generates the assets and metadata for physical formats (CD, vinyl, cassette) — track sequencing, disc/side layout, and barcode — for handoff to a manufacturing and retail partner.

Internet of Musical Things (@euterpe/iot)#

IoMusT (Internet of Musical Things) is the discipline of connecting physical instruments, sensors, and environments to digital music systems — enabling music creation through physical gesture, environmental data, and networked instrument collaboration.

Connected Instrument Network#

A device on the network moves through discovered → authenticated → paired → active → disconnected; only an active device sends or receives musical data.

• Device discovery and pairing: Automatic discovery of networked instruments and sensors on local and cloud networks, with secure authentication and a capability-negotiation handshake — a joining device declares what it can send (MIDI, OSC, audio, sensor data) and receive, and the network configures routing from that declaration rather than assuming a fixed device type.
• MIDI 2.0 over network: Transport of the MIDI 2.0 protocol over IP, carrying its higher-resolution controllers and per-note expression, with the latency budget kept low enough for remote playing to feel responsive.
• OSC transport: Open Sound Control transport for high-resolution, timestamp-accurate control data — each message carries a timestamp so a receiver can schedule it precisely rather than acting on arrival.
• Multi-device synchronization: Distributes a shared master clock to every device and measures each device's individual transport latency, offsetting per device so all instruments sound on the same beat despite differing network and hardware delays.

Gesture and Sensor Mapping#

• Motion data processing: Fuses accelerometer, gyroscope, and magnetometer streams into a stable three-dimensional orientation and motion estimate, then maps motion axes to musical parameters — a conductor's baton motion to tempo and dynamics, a dancer's body position to melody contour. Each mapping has a configurable input range and response curve so a physical gesture's scale matches the musical parameter's range.
• Gesture recognition: A library of recognized musical gestures — conductor beat patterns, air-instrument motions, dance-move vocabularies — each matched with a confidence; a recognized gesture above the confidence threshold triggers its bound music-generation parameter change.
• Biometric integration: Maps biometric streams to musical parameters — heart rate to tempo or intensity, breath sensors to phrase length, proximity sensors to interactive spatial position — within explicit safe input bounds so out-of-range sensor noise cannot drive an extreme musical result.
• Environmental sonification: Converts non-musical sensor data — weather, room occupancy, air quality, plant bioelectrical signals — into music by mapping each data dimension to a musical parameter (pitch, density, timbre) through a defined scaling, so the music tracks the environment meaningfully rather than arbitrarily.

Smart Studio Environment#

• Central device management: A single control surface for all connected studio devices, routing audio, MIDI, and control signals between any pair — the routing matrix is the authority for signal flow in the room.
• Intelligent automation: Schedules studio behaviors on time, occupancy, or musical context — lights dim when recording begins, reverb tail length adapts to measured room acoustics — as rules with a trigger condition and an action, evaluated as device state changes.
• Voice command interface: Hands-free control of connected devices during recording and performance, so transport and routing can be operated without touching a surface mid-take.

Autonomous Music AI Workflows (@euterpe/agents)#

AI agents that carry out multi-step music production and business tasks without human input at every decision. Each agent runs as a goal, a plan of steps, and an execution loop; an agent run moves through planned → executing → awaiting-review → completed | failed. Agents operate within explicit bounds — a step that spends money, publishes externally, or submits a release pauses at awaiting-review for human approval rather than acting unilaterally. Every action is logged so an autonomous run is fully auditable after the fact.

Release Management Agent#

• End-to-end release automation: Autonomous orchestration of the complete release process — collecting all assets, checking metadata, converting audio formats, submitting to distributors, setting up pre-save campaigns, generating social media announcements, and monitoring post-release performance.
• Quality control agent: Pre-release quality checks including audio loudness compliance, metadata completeness, cover art specification verification, and explicit content flagging.
• Post-release performance agent: Monitoring streaming, social media, and playlist performance after release and generating weekly performance reports with actionable recommendations.

A&R Scout Agent#

• Trend monitoring: Continuously tracking genre emergence, sound pattern virality, and geographic market opportunities across all major DSPs and social platforms.
• Emerging artist detection: Identifying rising artists based on velocity of growth rather than absolute numbers — catching breakout moments before they become mainstream.
• Collaboration matching: Identifying collaboration opportunities between artists based on sonic compatibility, audience overlap, and mutual career benefit.

Mix Engineer Agent#

• Automatic mixing: Autonomous mix engineering — analyzing all stems (drums, bass, guitars, vocals, synthesizers) and producing a professionally mixed stereo output calibrated to genre-appropriate reference tracks.
• Reference matching: Comparing the mix's spectral balance, dynamic range, stereo width, and loudness against user-specified reference tracks and automatically adjusting to match.
• Revision handling: Responding to human feedback in natural language ("the vocals are buried in the chorus," "the kick needs more punch") and applying targeted corrections.

Content Creation Agent#

• Social media content batch production: Generating a full week's worth of social media content from a single music release — short-form clips, lyric excerpts, behind-the-scenes concepts, and engagement-optimized captions for each platform.
• Royalty auditing agent: Automated cross-platform royalty reconciliation — collecting streaming data from all DSPs, comparing against expected payments, detecting discrepancies, and generating dispute claims for underpaid royalties.

Music Marketing and Promotion (@euterpe/marketing)#

AI-powered music promotion spanning organic social media strategy, paid advertising, press relations, and playlist pitching.

Social Media Content Generation#

• Short-form video creation: Extracting and formatting the most shareable moments from a song as TikTok, Instagram Reels, and YouTube Shorts clips — selecting the hook segment most likely to drive algorithmic amplification.
• Platform-native content: Generating content in formats native to each platform rather than one-size-fits-all posts — including animated cover art, lyric quote graphics, countdown teasers, and reaction compilation formats.
• Cross-platform adaptation: Automatically adapting a single piece of content to the correct dimensions, duration, and style requirements of each platform — a single generation produces variants optimized for every destination.

TikTok Viral Strategy#

• Viral potential scoring: Scores a song on the acoustic and structural features statistically correlated with TikTok virality — hook speed, BPM, energy curve in the first 3 seconds, and trend-compatibility — returning a 0–100 score with the per-feature contributions broken out, so a low score names which feature to address rather than just discouraging the track.
• Hook identification: Identifies the optimal 15-second clip for TikTok sound promotion, ranking candidate segments by engagement potential and returning the best clip's in/out timecodes.
• UGC campaign management: Designing and managing user-generated content campaigns where fans create videos using the artist's sound — the primary driver of organic viral spread on TikTok.
• Influencer matching: Identifying creators whose audience, content style, and engagement rates make them ideal partners for a specific track.

Paid Advertising Automation#

• Ad creative generation: Producing multiple ad creative variants for Facebook, Instagram, TikTok, YouTube, and Spotify Ad Studio from a single brief — enabling rapid A/B testing without manual design work.
• Audience targeting optimization: AI-driven audience targeting using lookalike audiences, interest stacking, and behavioral data to minimize cost-per-stream and cost-per-follower.
• Budget allocation intelligence: Distributing ad spend across platforms and creative variants to maximize reach within budget constraints, with real-time reallocation as performance data accumulates.

PR and Press Management#

• Press release generation: Producing publication-ready press releases for releases, tours, and milestones — following the inverted pyramid structure and journalistic conventions of the music press.
• Playlist pitching: Generating personalized pitch submissions to Spotify editorial playlist curators, Apple Music editors, and independent playlist curators — each pitch tailored to the specific playlist's aesthetic and audience.
• Media coverage tracking: Monitoring music publications, blogs, podcasts, and radio for mentions — aggregating coverage and calculating PR value.

Real-Time AI Accompaniment (@euterpe/accompany)#

AI-powered accompaniment systems that listen to a live performer and respond musically in real time — functioning as a full backing band that follows the human rather than the other way around.

AI Backing Band#

• Full rhythm section generation: Real-time generation of drums, bass, piano, and guitar accompaniment that follows the performer's chord input and tempo — producing a complete band feel from a single musician's performance.
• Style and genre adaptation: Switching between accompaniment styles (jazz trio, rock band, orchestral strings, reggae rhythm section, bossa nova combo) without stopping — the AI adapts its vocabulary, feel, and instrumentation to the selected genre.
• Dynamic fill and ending generation: Generating contextually appropriate fills between phrases and endings at the end of sections — behaviors that make the AI feel like a responsive musical partner rather than a loop playback system.
• Band-in-a-Box compatible interface: Integration with the industry-standard AI accompaniment workflow for musicians who already have established Band-in-a-Box workflows.

Adaptive Practice Accompaniment#

• Wait mode: The accompaniment pauses until the performer plays the next correct note — removing the pressure of tempo while learning new material.
• Tempo ramping: Gradually increasing accompaniment tempo across practice sessions to develop speed without the frustration of playing too fast too soon.
• Instrument isolation: Removing individual instruments from the backing (e.g., removing bass for bass practice, or removing piano for pianists learning to play over a rhythm section) so the performer hears exactly what their instrument adds.
• Performance feedback: Detecting missed notes, timing errors, and dynamic inconsistencies in real time and generating structured practice feedback.

Real-Time Jam Partner#

• Live audio analysis: Detecting the chord, key, scale, and tempo of whatever the human is playing in real time — enabling fully unscripted improvisation without any pre-programming.
• Complementary melody generation: Generating a responsive second voice that complements the human's melodic line rather than doubling it — creating genuine musical conversation.
• Call-and-response improvisation: Trading solos with the AI in jazz-style trading — human plays 4 bars, AI plays 4 bars, with each phrase responding to what came before.
• Low-latency processing: Audio analysis-to-response latency targeting sub-50ms for believable real-time musical interaction.

V2 Dynamic Music Ducking Context#

@euterpe/accompany supplies the dynamic music context for V2 commentary mix ducking: music intensity, genre preset, arrangement density, and active instrumentation. The bridge remains off-rollback and sends only audio-mix metadata to the Unreal V2.DynamicMusic bus.

Guitar and Amplifier Simulation (@euterpe/guitar)#

Neural network-based guitar and amplifier modeling reaching the fidelity of dedicated hardware solutions like the Kemper Profiler and Line 6 Helix — all within a software library.

Neural Amplifier Modeling#

• Amp capture pipeline: Training neural network models of any physical amplifier from a sweep of input/output audio pairs — capturing the non-linear behavior, sag, compression, and harmonic distortion that defines each amp's character.
• Model inference engine: Real-time inference of trained amp models at sub-5ms latency — mandatory for playable guitar monitoring without noticeable echo.
• Classic amp models: Pre-trained models of iconic amplifiers — Fender Twin Reverb (clean, glassy, articulate), Marshall Plexi (British crunch), Vox AC30 (chiming, chimey), Mesa/Boogie Dual Rectifier (high-gain American), and boutique models.
• NAM and AIDA-X format support: Compatibility with the two major open-source neural amp model formats — allowing users to share and import models from the growing community ecosystem of captured amps.

Pedal and Effects Chain#

• Drive pedal modeling: Neural or physics-based modeling of the most important overdrive, distortion, and fuzz pedals — Tube Screamer, Big Muff, Klon Centaur, Boss DS-1, and user-captured custom pedals.
• Full effects chain: Modeled versions of standard guitar effects: compressor, chorus, flanger, phaser, delay (tape/analog/digital), reverb, wah, tremolo, vibrato, octave divider, and pitch shifter.
• Pedalboard builder: Drag-and-drop signal chain construction with correct signal ordering (drive before modulation, modulation before reverb is standard; but pre-delay reverb and other creative orders are also supported).
• Cabinet and microphone simulation: Impulse Response (IR) convolution simulating microphone placement on speaker cabinets — choosing between a close-miked center-of-cone position and a brighter off-axis position changes the character significantly.

Guitar Intelligence#

• Chord and scale detection: Real-time detection of chords and scales being played from guitar audio input — displaying chord names, scale identification, and position on the fretboard.
• Tablature generation: Automatic guitar tab generation from audio input — producing playable tab notation rather than just MIDI.
• Technique detection: Identifying playing techniques (hammer-ons, pull-offs, bends, vibrato, slides, tapping, harmonics) from audio input for educational and analytical purposes.

Accessibility and Inclusive Music Design (@euterpe/access)#

Comprehensive accessibility infrastructure ensuring that Euterpe's music creation tools are usable by people with visual, hearing, motor, and cognitive disabilities.

Visual Accessibility#

• Full screen reader support: Complete ARIA labeling and landmark navigation so that every control, label, and status indicator in the DAW interface is accessible via NVDA, JAWS, VoiceOver, and other screen readers.
• Non-visual DAW workflow: A complete workflow for producing music without looking at the screen — audio descriptions of waveforms, spoken announcements of playhead position and clip boundaries, and sonification of visual data such as the mixer's frequency spectrum.
• Braille music notation export: Exporting notation in Braille music code for users who read music in Braille.
• Voice command interface: Hands-free control of all DAW functions — record, edit, and mix using spoken commands rather than keyboard and mouse.

Hearing Accessibility#

• Visual music representation: Rich visual representations of audio information — waveform displays, spectrum analyzers, beat-synchronized animations, and pitch display — enabling deaf and hard-of-hearing users to work with audio through vision.
• Haptic music feedback: Integration with haptic vests (SubPac) and wristband actuators — translating bass frequencies, kick drum hits, and rhythmic patterns into tactile physical sensation.
• Hearing aid and cochlear implant optimization: Frequency-specific processing modes that account for the different frequency response characteristics of hearing aids and cochlear implants, making the music clearer for those listeners.
• Visual metronome: A beat indicator that is prominently visible and customizable in size, color, and animation style — suitable for deaf musicians who track time visually.

Motor Accessibility#

• One-handed operation: All core functions operable with one hand, with configurable key mappings that avoid chords requiring two hands simultaneously.
• Switch control support: Full accessibility for users who navigate using a switch, scanning interface, or other single-point input device.
• Eye tracking input: Integration with eye tracking hardware (Tobii, built-in laptop cameras) for hands-free cursor control.
• Breath controller and foot pedal support: MIDI-mappable physical controls beyond keyboard and mouse — important for users who cannot use hands effectively.
• Xbox Adaptive Controller: Plug-and-play support for Microsoft's Adaptive Controller, which is designed to connect to a wide variety of adaptive switches and joysticks.

Cognitive Accessibility#

• Simplified interface mode: A stripped-back interface showing only essential controls — reducing cognitive load for users with cognitive disabilities, ADHD, or anxiety.
• Step-by-step guided workflows: Breaking complex production tasks into small, clearly described steps with visual progress tracking.
• WCAG 2.1 AAA compliance: Meeting the highest level of Web Content Accessibility Guidelines — ensuring minimum contrast ratios, no use of color as the sole information carrier, and no content that flashes more than three times per second.

Podcast and Audio Content Creation (@euterpe/podcast)#

AI-powered tools for creating professional podcast music, audio branding, and spoken-word audio content.

Podcast Music Generation#

• Intro and outro generation: Custom podcast theme music generated from a description of the show's topic, tone, and target audience — producing music that fits the podcast's identity rather than generic royalty-free tracks.
• Transition stingers: Short musical punctuation marks (2–5 seconds) used between segments — generated to match the energy and style of the main theme.
• Background underscore: Ambient music beds generated for specific segment types — interview beds, monologue underscore, sponsored segment music — each at a loudness calibrated for comfortable ducking under speech.
• Ducking-friendly design: All podcast music is generated with natural dynamic range that allows automatic ducking (volume reduction when speech is detected) without artifacts.

Jingle and Audio Branding#

• Sonic logo generation: The 2–5 second audio mnemonic that brands a company or show acoustically — the audio equivalent of a visual logo. Generated from brand personality descriptors and reference sonic identities.
• Radio ID and DJ drop generation: Station identification audio in multiple styles for radio stations, streaming channels, and podcast networks.
• Notification and UI sound design: Alert sounds, notification tones, and UI feedback sounds for apps and interactive products — maintaining consistent sonic brand identity across all touchpoints.
• Audio style guide: Generating a formal audio branding document specifying the key, tempo range, instrumentation palette, and sound character for all audio content produced by a brand.

Voice Integration#

• AI voiceover generation: Professional-grade AI voiceover for podcast intros, ad reads, and show segments — with style control ranging from conversational to broadcast-announcer.
• Ad read production: Producing complete sponsored segment audio from a script — combining AI voiceover with music beds and sound effects in a single production step.
• Multi-language podcast adaptation: Generating the same podcast intro and branding in multiple languages from a single brand brief — enabling international podcast distribution without re-recording.

Neith Audio Closure Dependencies (Phases 132, 133, 148, 167)#

Euterpe owns the music, composition, sonic intelligence, rights, and creative workflow domain. Neith owns the low-level audio runtime, DAW shell, DJ engine, and game-audio architecture. These two domains are complementary: Neith provides the professional hardware abstraction and plugin hosting that Euterpe's music intelligence runs on top of.

The four planned Neith integration phases and the Euterpe contributions to each are listed below.

• Sovereign audio runtime (Phase 132): Euterpe depends on @neith/audio-runtime, @neith/audio-graph, VST3/AU/LV2/CLAP hosting, MIDI 1.0/2.0, MPE, hardware controllers, clock/transport sync, network audio, and broadcast audio file I/O for professional-grade playback, recording, synthesis, plugin processing, and music-production validation. Euterpe's music-domain intelligence runs on top of this runtime layer.

• Composer DAW expansion (Phase 133): Euterpe supplies composition, transcription, theory, harmony, arrangement, scoring, mastering intelligence, stem generation, sample/preset metadata, and musical AI workflows. These sit on top of Neith's notation, clip launcher, restoration, pitch/time, mastering, mix assist, automation, comping, surround/immersive mixing, and deliverable tooling.

• DJ workstation (Phase 148): Euterpe contributes beat, key, phrase, genre, mood, stem, transition, crate, recommendation, licensing, and set-planning intelligence to Neith's deck engine, mixer, timecode control, hot cues, loops, sample decks, real-time stem isolation, FX, controller maps, recording, streaming, Ableton Link, MIDI clock, and show sync.

• Engine audio architecture (Phase 167): Euterpe feeds adaptive music, generative scoring, procedural SFX, dialogue/music ducking, V2 commentary mix ducking from V2.Commentary into V2.DynamicMusic, sound-brand generation, instrument/sound libraries, accessibility sonification, and music-rights metadata into Neith's source playback, 3D spatialization, acoustics, mixing, dialogue, music, synthesis, voice, profiler, tooling, and accessibility stacks.

Training-Data Flywheel (Phases 85–86)#

libs/euterpe/training-data implements this domain's side of the ML-sovereignty data flywheel: a training-data pipeline that captures music-creation interactions (compositions, arrangements, feedback loops) as passive training signals. Signals are consent-gated, anonymized where required, and emitted in the shared flywheel envelope that Nous dataset management (Phase 87) ingests for training and evaluation. Nous owns the training infrastructure; this domain owns what constitutes a high-quality domain signal.