whisper.cpp Transcription

Disclaimer

This specific page uses temporary drafts written and maintained by Generative AI. While verified to be accurate, please cross-verify with source code.

whisper.cpp can be used when its HTTP server exposes an OpenAI-compatible POST /v1/audio/transcriptions endpoint.

Setup

Start your whisper.cpp HTTP server and confirm it exposes an OpenAI-compatible transcription endpoint:

• POST /v1/audio/transcriptions
• GET /v1/models or GET /models

Keep the server running while TomoriBot is using it. The endpoint URL is the server root, such as http://127.0.0.1:8022.

If your whisper.cpp build exposes a different endpoint shape, place a thin wrapper in front of it that maps requests to TomoriBot’s expected OpenAI-compatible shape.

Register in TomoriBot

Run /provider custom-endpoint add:

• capability: transcription
• api_style: openai-compatible-transcription
• endpoint_url: your whisper.cpp server root

In the modal:

• Transcription Model: the model name your server reports
• Transcription Language: optional language hint, such as en or ja

Registration makes the endpoint active immediately. Use /model transcription later only when switching endpoints.

Use Transcripts

After registration, TomoriBot transcribes audio attachments in the background and adds the text to chat context. Use /speech transcripts only if you also want transcripts posted visibly in chat.