fast-whisper-diarizer 0.1.22

A package for audio transcription and speaker diarization using Whisper and NeMo toolkit

Fast Whisper Diarizer

Fast Whisper Diarizer is a Python package for audio transcription and speaker diarization using the Whisper model and NeMo toolkit.

Installation

To install the package, run:

pip install fast-whisper-diarizer

Initialization

Before using the diarize_audio function, you need to initialize the models using the initialize_models function. This ensures that the necessary models are loaded and ready for processing.

Example:

from fast_whisper_diarizer import initialize_models

whisper_model_name = "tiny.en"

# Initialize models
initialize_models(whisper_model_name)

Usage

To use the diarize_audio function, you can follow the example below. This function allows you to process audio data for transcription and speaker diarization, accepting both file paths and in-memory bytes data as input.

Example

from fast_whisper_diarizer import diarize_audio

# Example usage with a file path
final_transcript = diarize_audio(
    audio_data="path/to/audio/file.wav",
    whisper_model_name="tiny.en"
)
print(final_transcript)

# Example usage with in-memory bytes data
with open("path/to/audio/file.wav", "rb") as f:
    audio_bytes = f.read()

final_transcript = diarize_audio(
    audio_data=audio_bytes,
    whisper_model_name="tiny.en"
)
print(final_transcript)

Parameters

• audio_data (str or bytes): The input audio, either as a file path (str) or in-memory bytes data (bytes).
• whisper_model_name (str): The name of the Whisper model to use for transcription.

Optional Parameters

Parameter	Type	Default	Description
separate_vocals	bool	True	Whether to separate vocals from the audio.
processing_batch_size	int	8	Batch size for processing audio.
language_code	str	None	Language code for transcription.
suppress_numeric_tokens	bool	True	Whether to suppress numeric tokens in transcription.
computation_device	str	'cpu'	Device to use for computation ('cpu' or 'cuda').