Docker plus a supported NVIDIA GPU. The default 0.6B model uses about 4.6 GB of VRAM at startup; allow roughly 3.3 GiB more when loading optional timestamps.
Speech to text Stable v1.0 or rolling latest
Qwen3-ASR-STT
Private transcription for recordings and live speech.
A current local ASR service for 30-language transcription, bounded live microphone sessions, exact timestamps and subtitles, OpenAI-compatible clients, and opt-in MCP transcription from local files or URLs.
Open Transcribe, add an audio file or make a short recording, and select Transcribe. Use Stream for live microphone speech, API for integration status, and System for GPU and runtime controls.
The full image includes its ASR models and can transcribe without sending recordings to a hosted service after the image has been pulled.
What is included
A complete local workspace, not just a model.
- 30-language transcription
- Uploads, recording, and live microphone
- Bounded realtime sessions
- Exact timestamps, SRT, and VTT
- OpenAI-compatible API
- Latest: MCP transcription from files and URLs