Video to Transcript with Diarization

This project allows users to generate a transcript from a video file or a YouTube URL, with optional speaker annotation. The transcript is saved as an SRT or text file, which is widely used for subtitles.

Features

• Speaker annotation (Diarization): Supports multiple speakers and diarization for accurate transcription.
• Video and YouTube Support: Process local video files or YouTube URLs for transcription.
• FFmpeg Integration: Automatically converts MP4 files to MP3 for audio extraction.

Requirements

To get started, install the necessary dependencies:

pip install -r requirements.txt

Ensure you have ffmpeg installed on your system. For macOS, you can install it via Homebrew:

brew install ffmpeg

Configuration

Before running the script, update the config.ini file to include your credentials:

[WHISPER]
model = distil-large-v3

[PYANNOTE]
auth_token = YOUR_HUGGING_FACE_TOKEN

[OPENAI]
api_key = YOUR_OPENAI_API_KEY

• WHISPER model: Specifies the model used for transcription.
• PYANNOTE auth_token: Replace with your Hugging Face token to enable speaker diarization.
• OPENAI api_key: Add your OpenAI API key for Whisper integration.

How to Use

1. Run the script:

   python main.py

2. Input type:

   • You will be prompted to choose between a YouTube URL (1) or a local file (2).
   • If choosing a file, provide the path to an MP4 or MP3 file.

3. Diarization:

   • Choose y or n to enable or disable speaker diarization.

4. Output:

   • The transcription will be saved in the transcripts/ folder with a timestamped filename.

Example Usage

• For a YouTube URL:

  python main.py
  Enter input type, [1] for YouTube URL, [2] for filepath: 1
  Enter URL: https://youtube.com/example
  Enter filename: example_transcript
  Enable diarization for multiple speakers? [y/n]: n

• For a local MP4 file:

  python main.py
  Enter input type, [1] for YouTube URL, [2] for filepath: 2
  Enter filepath: /path/to/video.mp4
  Enable diarization for multiple speakers? [y/n]: y

Output Example

Below is a sample output from a transcription run:

[
  {
    "start": 6.411,
    "end": 33.459,
    "text": " Everybody's asleep, man, or getting home after a long night..."
  },
  {
    "start": 33.883,
    "end": 58.316,
    "text": " What's not always open is the opportunity to check the box in life..."
  },
  {
    "start": 58.774,
    "end": 88.303,
    "text": " Everyone else is probably getting home from some party right now..."
  },
  ...
]

The full transcription can be found in the transcripts/ folder. For example, transcripts/{input_filename}_{timestamp}.txt

Notes

• The script converts MP4 files to MP3 for audio extraction using ffmpeg.
• Ensure your Hugging Face and OpenAI API keys are correct in config.ini to enable full functionality.
• The output is saved in the transcripts/ directory in a timestamped text file.

License

This project is licensed under the MIT License. See the LICENSE file for more details.