transcribe

To transcribe audio with speaker diarization, provide an audio file and optional known-speaker hints, then run the transcription to output a diarized JSON format. It uses the gpt-4o-transcribe-diarize model to separate speakers automatically.

You can transcribe audio files larger than 25 MB because the process includes auto chunking for longer clips. It validates the OPENAI_API_KEY and processes the segments sequentially to bypass the 25 MB per-file limit.

Supported transcription output formats include plain text, standard JSON, and diarized JSON. These formats accommodate meetings and interviews by allowing optional known-speaker references to accurately identify participants in the written output.

You need an OpenAI API key to transcribe audio to text. The process validates the OPENAI_API_KEY before execution and relies on OpenAI models like gpt-4o-mini-transcribe to generate deterministic text results.

Known-speaker diarization works by accepting optional speaker hints alongside the audio file, then using the gpt-4o-transcribe-diarize model to assign dialogue accurately. This produces a diarized JSON output that labels participants by name.

The default model for speech-to-text transcription is gpt-4o-mini-transcribe. It provides deterministic results for standard audio inputs, while gpt-4o-transcribe-diarize is specifically applied when speaker diarization is required.