transcribe

To transcribe audio with speaker identification, provide known-speaker hints to enable the diarization model. The CLI uses the gpt-4o-transcribe-diarize model to separate and label speakers, outputting results in a diarized_json format.

The best way to extract text from meeting recordings is using a bundled CLI with the gpt-4o-mini-transcribe model. It transcribes speech into text and saves the transcripts directly to an output directory for deterministic, repeatable processing.

Yes, you can use OpenAI transcription models for podcast audio processing. The CLI supports OpenAI dependencies and applies default models like gpt-4o-mini-transcribe to convert spoken podcast content into accurate text transcripts.

Audio diarization requires known-speaker hints to function properly. When you provide speaker hints, the CLI switches to the gpt-4o-transcribe-diarize model; without them, it defaults to standard text output without speaker labeling.

When transcribing interviews, the supported output formats are text, json, and diarized_json. You can direct the transcript to a specific file using --out, to a directory using --out-dir, or print directly to --stdout.