Xircuits Audio Transcription Components

A Xircuits component library for transcribing audio into text with speaker diarization. This library provides components for:

• Loading and processing audio files
• Performing speaker diarization (identifying who spoke when)
• Transcribing speech to text
• Combining diarization and transcription results
• Saving formatted transcripts

Prerequisites

This component library requires:

1. Access to the Hugging Face Hub models:

   • You need to accept the terms of use for the pyannote models at:
     • https://huggingface.co/pyannote/speaker-diarization
     • https://huggingface.co/pyannote/segmentation
   • A Hugging Face access token for the diarization models

2. Sufficient disk space for the downloaded models (approximately 1-2GB)

Installation

To use this component library, ensure you have Xircuits installed, then simply run:

xircuits install https://github.com/xpressai/xai-transcribe

Alternatively you may manually copy the directory / clone or submodule the repository to your working Xircuits project directory then install the packages using:

pip install -r requirements.txt

Usage

The library provides components for a complete audio transcription pipeline:

1. TranscribeLoadAudioFile - Load an audio file or use a sample dataset
2. TranscribeSpeakerDiarization - Identify different speakers in the audio
3. TranscribeSpeechTranscription - Transcribe the audio to text with timestamps
4. TranscribeCombineDiarizationAndTranscription - Combine speaker information with transcription
5. TranscribeSaveTranscriptToFile - Save the formatted transcript to a file

Example

Create a new Xircuits workflow and add the components in sequence:

1. Start with TranscribeLoadAudioFile and provide a path to your audio file
2. Connect to TranscribeSpeakerDiarization (set use_auth_token to True if using Hugging Face models)
3. Add TranscribeSpeechTranscription (defaults to Whisper base model)
4. Connect both to TranscribeCombineDiarizationAndTranscription
5. Finally connect to TranscribeSaveTranscriptToFile to save the results

Tests

A github action to test your workflow runs has been provided. Simply add the path of your workflows here.