README.md

metadata

title: Soprano TTS - Single File
emoji: 🎧
colorFrom: yellow
colorTo: purple
sdk: static
short_description: Single self-contained HTML file for browser TTS
app_file: index.html
pinned: false
license: apache-2.0
custom_headers:
  cross-origin-embedder-policy: require-corp
  cross-origin-opener-policy: same-origin
  cross-origin-resource-policy: cross-origin

Soprano TTS — Single Self-Contained File

Single-file version by Nekochu · Soprano by ekwek1 · ONNX port by KevinAHM · Apache-2.0

---

A single HTML file (~32KB) that provides real-time neural text-to-speech in the browser.

All code, styles, and logic are embedded in one file. Just open index.html and start using it.

---

✨ Features

• 🚀 Single File: Everything in one HTML file - no dependencies, no build process
• 🎯 Fast Models: Uses INT8 quantized ONNX models for optimal performance
• 🔊 Real-Time Streaming: Web Worker + AudioWorklet for smooth playback
• 📊 Live Metrics: Shows TTFB and Real-Time Factor
• ☁️ CDN Models: Loads models directly from HuggingFace

---

🚀 Usage

Option 1: Direct Browser

Simply open index.html in a modern browser (Chrome, Edge, Firefox).

Option 2: Local Server

python -m http.server 8000

Then visit http://localhost:8000

---

📝 How It Works

1. Models (~112MB total) are loaded from HuggingFace CDN on first use
2. Text preprocessing and ONNX inference runs in a Web Worker
3. Audio chunks stream to AudioWorklet for smooth playback
4. Browser caches models after first load

---

🎯 Models Used

• Backbone: soprano_backbone_kv_int8.onnx (80.9 MB)
• Decoder: soprano_decoder_int8.onnx (30.8 MB)
• Tokenizer: Loaded from HuggingFace Transformers.js

All models load directly from HuggingFace CDN - no local files needed!

---

💡 Credits

• Original Soprano: ekwek1/soprano
• ONNX Port: KevinAHM/soprano-web-onnx
• Combined into one HTML: Nekochu (this space)

---

📜 License

Apache-2.0 (same as upstream Soprano)