Voice Presentation Analyzer (VPA)

Languages: English | 繁體中文 | 簡體中文 | 日本語

---

A 100% privacy-first, in-browser voice impression and acoustic analysis tool. Evaluates feminine / masculine voice impression tendencies, provides quick test challenges, practice drawers, real-time acoustic pitch & formant tracking, dynamic voice card sharing, and advanced analytics.

• 🌐 Live Demo (GitHub Pages): https://shusei.github.io/vpa
• 📋 Changelog: CHANGELOG.md
• 🔬 Algorithm Report: ALGORITHM_VERIFICATION.md
• 🔒 Privacy Guarantee: Audio processing and neural model inference are 100% local inside your browser. No voice data is ever uploaded to any server.
• 🌍 Multilingual Support: English, Traditional Chinese (繁體中文), Simplified Chinese (簡體中文), and Japanese (日本語) with automatic browser locale detection.

---

Table of Contents

• Key Features & Dual Modes
  • 1. Quick Test Experience
  • 2. Professional Experience
  • 3. Practice Drawer
  • 4. Dynamic Social Card Sharing
• Core Capabilities
• Workflow & Usage
• Technical Architecture
• Development & Testing
• Contributing
• Privacy & Disclaimer
• License & Credits

---

Key Features & Dual Modes

Voice Presentation Analyzer (VPA) offers dual interface modes tailored for both quick daily voice check-ins and deep acoustic analysis:

1. Quick Test Experience

• Daily Test: Read a short daily phrase to get a voice impression score in seconds.
• Standard Challenge: Read 3 consecutive prompts to measure pitch stability and consistency across sentences.
• Score Card & Replay Controls: Shows impression percentage, estimated voice age, and archetype. Replay buttons support Play/Pause toggle.

2. Professional Experience

• Real-time Pitch Stream: Live 50–450 Hz pitch contour, instantaneous loudness, and noise floor monitoring.
• Formant & Resonance Panel: Real-time estimation of F1–F3 formants, chest / mask / head resonance proportions, and breathiness ratio.
• Intonation & Advanced Analytics: Intonation curve visualization, speaking rate (syllables/sec), continuous voicing ratio, and personalized advice.

3. Practice Drawer

• Core-36 Phrase Library: Categorized phrase drills with quick recording, instant replay, score tracking, and history comparison.

4. Dynamic Social Card Sharing

• Dynamic Voice Cards (PNG / Video)：Generate 9:16 video cards or custom PNG score graphics completely inside the browser.
• One-Tap Social Sharing: Direct short-link and card sharing to X (Twitter), Threads, and LINE.

---

Core Capabilities

Privacy & Performance

• Local Inference: Built with @xenova/transformers and ONNX Runtime (WebGPU / WASM). Models run locally once cached in IndexedDB.
• Offline Support: Works 100% offline after the initial model download.
• Format Support: Supports MediaRecorder live recording and file uploads (.mp3, .m4a, .mp4, .mov, .wav).

---

Development & Testing

npm test      # Run unit tests, syntax checks, and markup checks
npm run test:e2e # Run Playwright end-to-end browser test suite

---

Contributing

Bug fixes, translation fixes, and documentation improvements are welcome. See the Contributing Guide for the fork, branch, test, commit, push, and Pull Request workflow.

---

Privacy & Disclaimer

• Non-Medical Tool: VPA is an acoustic feedback tool for voice exploration and training. It does not determine gender identity, legal status, or medical diagnoses.
• Vocal Safety: Practice at a comfortable volume. If you experience vocal fatigue or hoarseness, pause and consult a licensed speech-language pathologist or ENT physician.

---

License

• Project: MIT License
• Model: Apache-2.0 License (prithivMLmods/Common-Voice-Gender-Detection-ONNX)