speaches-ai/speaches

Health v2 · maintenance only

Adoption not part of the score

3689 stars · 455 forks observed · 2026-09-29

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Speaches is an OpenAI API-compatible self-hosted server for speech-to-text (via faster-whisper), translation, and text-to-speech (via Kokoro and Piper), with streaming and a realtime API. It aims to be 'Ollama for TTS/STT models', with dynamic model loading/offloading and Docker deployment.

Use cases

• self-host an OpenAI-compatible whisper transcription server
• stream transcribe audio files via API
• convert text to speech with kokoro or piper models
• translate audio to English using whisper
• build realtime voice assistant with speech-to-speech API
• replace OpenAI audio endpoints with local models
• deploy GPU or CPU speech server with docker compose

When to choose

• you want OpenAI API compatibility but self-hosted speech models
• you need streaming transcription or a realtime audio API
• you want dynamic model loading across multiple whisper/TTS models
• you want GPU and CPU support with Docker deployment

When to avoid

• you need the absolute highest-accuracy transcription from a managed cloud service
• you want a GUI desktop transcription app rather than an API server
• you need voice cloning or custom TTS voices beyond Kokoro/Piper

Facets

service · maturity active

speech-recognition tts http-server llm-inference speech-processing artificial-intelligence self-hosted apis self-hosted python whisper faster-whisper kokoro piper openai-api-compatible transcription translation streaming realtime-api audio docker linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="speaches-ai/speaches")