UTAU.js

Parametric singing voice synthesis in the browser and Node.js. Implements a source-filter model with an LF glottal source and cascaded formant filter.

Visit the live demo!

npm install utaujs

Features

• LF Glottal Source - Liljencrants-Fant pulse model with independent control of OQ, SQ, tenseness, aspiration noise, jitter, and shimmer
• Formant Cascade - Series of resonators and anti-resonators modelling the vocal-tract transfer function, with smooth cross-phoneme interpolation
• Three Languages - English (ARPAbet + CMUDict G2P), Japanese (hiragana/ romaji + kanji), Mandarin (pinyin decomposition). Extensible via registerLanguage
• File Import / Export - 16 import + 14 export formats from UTAU, VOCALOID, Synthesizer V, CeVIO, MusicXML, Standard MIDI, and more via utaformatix-ts
• Streaming Playback - AsyncGenerator-based render pipeline with Web Audio batch scheduling and buffer pooling
• OneVoice System - One parametric voice config tuned to perceptual dimensions. scaleVoice() maps high-level sliders (gender, breathiness, brightness, tension) to the underlying DSP parameters
• Zero Native Dependencies - Pure TypeScript, runs anywhere Node.js 18+ or a modern browser runs

Quickstart

import { renderScore, femaleVoice, japanese, encodeWav } from "utaujs";
import { writeFileSync } from "node:fs";

const score = {
  tempos: [{ tick: 0, tempo: 120 }],
  resolution: 480,
  notes: [
    { lyric: "こ", noteNum: 60, length: 480 },
    { lyric: "ん", noteNum: 62, length: 480 },
    { lyric: "に", noteNum: 64, length: 480 },
    { lyric: "ち", noteNum: 65, length: 480 },
    { lyric: "は", noteNum: 67, length: 960 },
  ],
};

const chunks = await renderScore(score, femaleVoice, "jp");
const wav = encodeWav(chunks);
writeFileSync("hello.wav", Buffer.from(wav));

Stream to audio output in the browser:

import { streamScore, femaleVoice, japanese, StreamPlayer } from "utaujs";

const player = new StreamPlayer();
await player.play(streamScore(score, femaleVoice, "jp"));

API Overview

| Export | Description | |--------------------------------------------------------------|----------------------------------------------| | renderScore | Render full score to AudioChunk[] | | streamScore | Render score as AsyncGenerator for streaming | | encodeWav | Encode AudioChunk[] to WAV bytes | | mixChunks | Mix overlapping chunks into a single buffer | | buildVoice / scaleVoice | Create and parameterize VoiceConfig presets | | maleVoice / femaleVoice | Built-in voice presets | | japanese / english / mandarin | Language modules with G2P | | importScoreFromFile / importScoreFromBytes | Import file formats | | exportScoreToBytes / exportScoreToBlob / downloadScore | Export to any format | | scoreToUfData / ufDataToScore | Low-level UtaFormatix data conversion | | StreamPlayer | Web Audio playback with real-time scheduling | | registerLanguage / registerVoice | Extend with custom languages or voices |

Full API reference at Docs/ or https://nellowtcs.me/UTAU.js/docs.

Architecture

mindmap
  root((UTAU.js))
    Core DSP
      LFGlottalSource
      FormantCascade
      NoiseSource
      Envelope / Mix
    Languages
      Japanese G2P
      English CMUDict
      Mandarin Pinyin
    Synth
      Renderer
      Streamer
      Chunk Mixer
    Voices
      Presets (male, female)
      Builder / Scaler
    IO
      Import (utaformatix-ts)
      Export (utaformatix-ts)
      WAV encoder

Development

# clone with submodules (required for CMUDict)
git clone https://github.com/NellowTCS/UTAU.js.git --recurse-submodules
cd UTAU.js/Build

npm install
npm run build     # builds G2P data + library
npm test          # 160+ tests via Jest
npm run typecheck # tsc --noEmit
npm run lint      # ESLint

The docs site lives in Docs/ and uses DocMD. To preview:

cd Docs
npm install
npm run dev

License

MIT - see LICENSE.