transcribe

You can transcribe meeting audio to text with speaker labels by using Whisper for transcription and optional pyannote.audio for diarization. It processes common formats like .m4a, .mp3, and .wav, saving transcripts with labeled speakers.

Yes, you can automatically detect and transcribe new audio files in a folder. Running the transcription command on a directory will identify untranscribed audio files and save the resulting text transcripts to a designated output folder.

You do not need an HF_TOKEN for standard Whisper transcription, but an optional HF_TOKEN is required for speaker diarization. The diarization feature uses pyannote.audio, which needs this authentication token to function.

The best way to extract action items after converting meeting audio to a transcript is to run a summarization command on the generated text file. This downstream workflow captures decisions and action items from the transcribed meeting content.

Yes, faster-whisper supports word-level timestamps for audio recordings. It converts your meeting audio into accurate text transcripts while providing precise timing data for each spoken word.

Audio diarization and transcription require Python dependencies including faster-whisper, torch, torchaudio, pyannote.audio, numpy, and tqdm. These libraries handle the speech-to-text processing, speaker identification, and audio data manipulation.