speaches

Model Serving & MLOps

OpenAI API-compatible server for streaming transcription, translation, and speech generation

Python Latest v0.9.0-rc.3 · 5mo ago Security brief →

Features

OpenAI API-compatible server for transcription, translation, and speech generation
Streaming support for real‑time transcription and speech‑to‑speech interactions
Dynamic model loading with GPU and CPU support

View all 6 releases →

v0.9.0-rc.3 New feature 5mo

⚠ Upgrade required

The `onnx-export` package has been removed; users relying on it should plan accordingly.
`huggingface_hub.list_models` deprecated parameters have been addressed — verify any custom integrations using this API.
Models are now downloaded automatically on startup; ensure sufficient disk space and network access in deployment environments.

Notable features

v0.9.0-rc.2 New feature 6mo

⚠ Upgrade required

This is a release candidate (rc.2); users should treat it as pre-release and validate before production use.

v0.8.3 Bug fix 8mo

Fixed corrupted audio headers for WAV and FLAC streaming, improved Piper model validation, and added CA certificates to Docker image.

v0.8.2 New feature 8mo

⚠ Upgrade required

VAD filter is now enabled by default for STT endpoints. Users who relied on previous behavior (VAD filter off) should explicitly set VAD_FILTER=False in their configuration to preserve existing transcription behavior.

Notable features