Self-hosted alternatives to openai/transcribe

O

openai/transcribe

Transcribe audio files to text with optional speaker diarization

transcribe is an agent skill from OpenAI's curated skills collection that turns audio files into text with OpenAI's transcription models. It produces plain text, and it can also do speaker diarization, which separates the speakers, with known-speaker hints for labeled transcripts. It works through a bundled Python command-line tool with automatic chunking for long files and configurable output formats. Using the bundled tool with automatic chunking handles a long file and multi-speaker labels in a single command, without splitting the audio by hand or fixing speaker attribution afterward. Typical uses include transcribing a recorded interview with labeled speaker turns, extracting text from a meeting recording saved as an .m4a file, converting a voice memo to plain text for editing, running diarization on a podcast to separate two known hosts, and batch transcribing several audio files into separate output folders. It suits journalists, researchers, students and anyone who works from recordings. Transcripts can contain errors, and speaker labels are only as good as the audio, so check important passages against the recording. Recordings often contain private conversations, so handle the files and the transcripts with appropriate care. The skill is listed on officialskills.sh and installs from the openai/skills repository. Install it with npx skills add https://github.com/openai/skills --skill transcribe.