Speaches - OpenAI-Compatible TTS/STT Server

speaches is an OpenAI API-compatible server supporting streaming transcription, translation, and speech generation. This project aims to be Ollama for TTS/STT models - making it easy to run multiple speech models with a unified API.

Powered By

• Speech-to-Text: faster-whisper
• Text-to-Speech:
  • Kokoro - #1 in TTS Arena
  • StyleTTS2 - High-quality neural TTS with voice cloning
  • Piper - Fast, lightweight TTS

For full documentation, visit: speaches.ai

Features

Core Capabilities

• OpenAI API Compatible - All tools and SDKs that work with OpenAI's API work with Speaches
• Streaming Support - Real-time transcription via SSE (no waiting for full audio processing)
• Dynamic Model Management - Automatic model loading/unloading based on usage
• GPU & CPU Support - Run on your preferred hardware
• Docker Ready - Easy deployment via Docker Compose
• Highly Configurable - Extensive configuration options

Audio Generation (Realtime API)

• Text → Audio: Generate spoken summaries
• Audio → Text: Transcription and sentiment analysis
• Audio → Audio: Speech-to-speech model interactions

Text-to-Speech Engines

• Kokoro - Ranked #1 in the TTS Arena
• StyleTTS2 - 24kHz high-quality synthesis, voice cloning, 18+ languages, 14 preset voices
• Piper - Fast and lightweight TTS

Demos

Realtime API

2025-05-29_21-40-00.webm

(Excuse the breathing lol. Didn't have enough time to record a better demo)

Streaming Transcription

TODO

Speech Generation

2025-01-12_13-20-58.webm

Quick Start

Text-to-Speech with StyleTTS2

curl -X POST http://localhost:8000/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{
    "model": "jakezp/StyleTTS2-LibriTTS",
    "input": "Hello world! This is high-quality neural speech synthesis.",
    "voice": "female_anna"
  }' \
  --output speech.wav

Available Voices: female_anna, male_gavin, female_emily, male_thomas, and 10 more preset voices.

Voice Cloning: Use "voice": "file:///path/to/reference.wav" to clone any voice.

Custom Parameters: Adjust alpha (0-1), beta (0-1), and diffusion_steps (1-20) for fine control.

Attribution

This project builds upon and integrates several excellent open-source projects:

• Speaches - The foundation of this OpenAI-compatible TTS/STT server (MIT License)
• StyleTTS2 - High-quality neural text-to-speech synthesis with style modeling (MIT License)
• StyleTTS2 HuggingFace Space - Reference implementation and inspiration for integration
• faster-whisper - Fast speech-to-text transcription
• Kokoro - State-of-the-art TTS model
• Piper - Lightweight TTS engine

License

This project is a combination and integration of the above works. All original projects retain their respective licenses (primarily MIT License). This derivative work is provided as-is for research and development purposes.

We express our gratitude to all the original authors and contributors of these projects. Their excellent work makes this integration possible.

Support & Issues

Please create an issue if you:

• Find a bug
• Have a question
• Want to suggest a feature

Contributing

Contributions are welcome! Please feel free to submit issues or pull requests.

---

Note: This is a community integration project combining multiple open-source TTS/STT engines into a unified OpenAI-compatible API. It is not affiliated with OpenAI.