transcribe

Transcribe recordings including speaker recognition (diarization) and timestamps.

Powered by OpenAI Whisper and pyannote-audio.

Setup

Prerequisites

• Python 3.11+ (Tested with 3.11.5)
• Hugging Face account

Steps

1. Install via pip:

   pip install git+https://github.com/sueskind/transcribe

2. Accept pyannote/segmentation's user conditions
3. Accept pyannote/speaker-diarization's user conditions
4. Create a Hugging Face access token

Usage

Example:

$ transcribe pulpfiction.mp3 -t <token>
Speaker 1 (00:00:00):
They don't call it a quarter pounder with cheese?

Speaker 2 (00:00:06):
No, they got the metric system there, they wouldn't know what a quarter pounder is.

Speaker 1 (00:00:13):
What do they call it?

Speaker 2 (00:00:17):
Royale with cheese.

Use --help to show all options:

$  transcribe --help
Usage: transcribe [OPTIONS] AUDIO

  Transcribe and diarize (recognize speakers) recorded audio.

Arguments:
  AUDIO  Path to the input audio file.  [required]

Options:
  -t, --token TEXT  Hugging face access token.  [required]
  -o, --out TEXT    Path to the output file. Print to stdout if not set.
  --device TEXT     Device to run the models on.  [default: cuda]
  --language TEXT   Spoken language in the recording.  [default: en]
  --help            Show this message and exit.