Summary
One tool, and a base URL that decides whose servers hear the recording.
transcribe_audio is the whole surface, which is the right size for a job with one input and one output. The endpoint address and the model name are the interesting options: point them at a local or self-hosted compatible endpoint and the audio never reaches OpenAI.
What it is
A small Node server with one tool: transcribe_audio takes the path to an audio file and returns the transcript.
What you get
- transcribe_audio, taking a filepath as its one required parameter
- An optional language hint as an ISO-639-1 code, for example en or es
- An option to write the transcription out to a file rather than returning it only in the conversation
- A configurable base URL and model, so it can point at an OpenAI-compatible endpoint instead of OpenAI
Requirements
Node, and a clone of the repository built with npm. An OpenAI API key in the environment; the endpoint address and the model name are optional overrides.