CrowCrow: Parametric Articulatory Voice Synth (Experimental)

CrowCrow main UI (click to enlarge)

---

CrowCrow is an experimental, browser-based parametric articulatory speech synthesizer.
This repo is not under active development, but is preserved as an artifact for reference, inspiration, or further research.

Note:
Like many experimental projects, some paths lead you nowhere but inform your future work.
CrowCrow began as an attempt to build a fully parametric TTS synth that could be AI-driven for future embodiment, as an alternative to current dataset/diffusion-based TTS.
However, since the synth itself is not the main focus and lacks essential features like a phonemizer, coarticulation, and prosody (which are better served with a different architecture), a full rewrite would be needed for production use.

Companion Post:
Embodied speech for AIs — A short introduction to the ideas behind this project.

Features

• Real-time articulatory speech synthesis in the browser (Web Audio API + AudioWorklet)
• Modular, extensible audio engine (Glottis, Tract, Nasal, Click, Tap, Trill, etc.)
• Partial IPA phoneme coverage (vowels, plosives, fricatives, nasals, clicks, trills, taps, etc.)
• Interactive UI: phoneme buttons, tract visualizer, oscilloscope, parameter sliders
• Phoneme text input and sequencer for custom utterances
• Designed for research, education, and creative coding

---

Live Demo

👉 Try CrowCrow on GitHub Pages

---

Screenshots

• Full UI:
  

• Tract Visualizer & Oscilloscope:
  

• Phoneme Text Input:
  

---

Getting Started

Online

Just visit: https://kenoleon.github.io/crowcrow/

Local

git clone https://github.com/kenoleon/crowcrow.git
cd crowcrow
npm install
npm run build
npm start
# Then open http://localhost:8080/

---

Usage

• Start Synth else you wont hear a thing.
• Click phoneme buttons to hear individual sounds.
• Move the master Gain to hear a continuous waveform.
• Adjust sliders for pitch, tenseness, tract shape, etc.
• Enter a sequence in the phoneme text input and press "Play" to synthesize custom utterances.
• Watch the tract visualizer and oscilloscope update in real time.

---

Architecture

• Audio Graph Topology

[subglottalNode]
      │
[noiseNode]─────┐
      │         │
  [glottisNode] │
      │         │
[chestinessNode]│
      │         │
[transientNode] │
   │     │      │
   │     └─────────────┐
   │                   │
[tractNode]        [nasalNode]
   │                   │
   └─────┬─────┬───────┘
         │     │
  [clickNode]  │
  [tapNode]    │
  [trillNode]  │
         │     │
    [summingNode]
          │
      [gainNode]
          │
  [masterGainNode]
          │
    [analyserNode]
          │
[audioContext.destination]

Legend:

• subglottalNode: Simulates subglottal pressure (lungs)

• noiseNode: Generates noise for fricatives, aspiration, etc.

• glottisNode: Simulates vocal fold vibration and voicing

• chestinessNode: Adds chest resonance

• transientNode: Handles plosive bursts and transients

• tractNode: Main vocal tract filter (8 zones, 44 segments)

• nasalNode: Simulates nasal tract coupling

• clickNode, tapNode, trillNode: Special consonant bursts

• summingNode: Sums oral, nasal, click, tap, trill outputs

• gainNode, masterGainNode: Output gain control

• analyserNode: For oscilloscope/visualization

• Processors:

  • public/audio/processors/: AudioWorkletProcessors for each stage
  • src/main.js: Main controller, node creation, UI, parameter flow
  • src/state/phonemeMap.js: Phoneme definitions and parameters

---

Phoneme Map

• All phonemes are defined in src/state/phonemeMap.js using articulator parameters.
• You can add or modify phonemes by editing this file and the UI.

---

Development

• Build: npm run build
• Dev server: npm start
• Deploy: Push to main branch; GitHub Actions will auto-deploy to Pages.

---

License

MIT License (see LICENSE)

---

Credits

• Created by Keno Leon
• Inspired by Pink Trombone and other open-source speech synthesis projects

---

Status

This repo is experimental and will not be developed further.
It is preserved as a reference artifact for the community.

---

Roadmap / Ideas (Not Planned)

• Coarticulation and expressive prosody
• Voice customization and presets
• AI-driven parameterization
• Improved UI/UX and accessibility

---

Feel free to fork or reference for your own projects!