itsnotgunnar / speakerdiarization

Transcription and speaker diarization of youtube videos. This will be merged into a larger repository in an effort to automate the process of creating personalized models for users.

speakerDiarization

This Python script contains a class AudioProcessor which is used for processing audio files.

Dependencies

This repository uses a conda environment - once you have Miniconda or Anaconda installed run the following script, replacing myenv with the name of your environment:

conda env create -f environment.yml
conda activate myenv
source activate myenv

Usage

Here's a simple example to get you started:

python

from AudioProcessor import AudioProcessor

Initialize the processor

processor = AudioProcessor(sounds_id="some_id", file_path="path/to/audio/file")

Run the main loop

processor.mainLoop()

Methods

init(self, sounds_id, file_path)

Initializes the AudioProcessor object. get_video_id(self, video_url)

Extracts the video ID from a given URL. speaker_diarization(self)

Performs speaker diarization on the audio file. millisec(self, timeStr)

Converts a time string to milliseconds. primary_speaker(self)

Identifies the primary speaker in the audio. chunk_primary(self)

Segments the audio based on the primary speaker. transcribe_directory(self)

Transcribes all .wav files in a given directory. audio_to_audioSeg(self, t1, t2)

Extracts an audio segment between t1 and t2. mainLoop(self)

The main function that ties all the methods together.

Environment Variables

The script uses the following environment variables:

• OPENAI_API_KEY: The API key for OpenAI.
• HUG_USER1: The API key for Hugging Face.

Make sure to set these environment variables before running the script.

Contributing

Feel free to open issues or PRs if you find any problems or have suggestions for improvements.