Audio Transcriber

An audio transcription tool leveraging OpenAI Whisper and WhisperX for accurate and efficient transcriptions.

Table of Contents

• About
• Features
• Prerequisites
• Installation
• Usage
• Configuration
• Project Structure
• Contributing
• License
• Author

About

Audio Transcriber is a Python-based command-line and scriptable tool for transcribing audio files into text. It integrates:

• OpenAI Whisper for baseline transcription.
• WhisperX for forced alignment and improved timestamp accuracy.
• Optional usage of DeepSeek or other transcription services.

The tool supports MP3 and MP4 input formats and outputs verbatim transcripts with timestamps.

Features

• High accuracy with WhisperX forced alignment.

• Timestamped transcriptions.

• Batch processing of multiple audio files.

• Simple CLI and Python API.

• Customizable model selection and performance parameters.

• This repository contains an app built using Gradio that inputs audio and video files and creates a transcription. The steps to create a transcription using this app are...

Create a new transcription:

1. Select the "Transcribe" tab
2. Enter the language spoken in the audio or video
3. Select an audio or video file on your local drive to upload
4. Select "Submit" to load the audio followed by "Transcribe" to begin the transcription
5. The results are displayed in the UI and written to an SQlite database.

View Available Transcriptions:

1. Select "View Transcriptions" tab
2. From the dropdown select the title of an audio or video
3. The transcripts results are displayed

Prerequisites

• Python 3.8+
• pip package manager
• FFmpeg installed and available in your PATH.

Installation

git clone https://github.com/krpopkin/Audio_Transcriber.git
cd Audio_Transcriber
pip install -r requirements.txt

Usage

Command-Line

python transcriber1.py --input /path/to/audio.mp3 --output transcript.txt

Common options:

• --model: Whisper model to use (e.g., base, small, medium, large).
• --align: Enable WhisperX forced alignment.
• --language: Specify language code (e.g., en, es).
• --batch-dir: Process all audio files in a directory.

Python API

from transcriber1 import transcribe

result = transcribe(
    input_path="audio.mp3",
    model="small",
    align=True,
    language="en"
)
print(result.text)

Configuration

Configuration options are set via command-line flags or environment variables:

• WHISPER_MODEL
• USE_WHISPERX
• FFMPEG_PATH

Project Structure

Audio_Transcriber/
├── transcriber1.py           # Main transcription script
├── transcription_service.py  # Whisper and WhisperX integration
├── requirements.txt
├── examples/                 # Example audio files and usage
│   └── sample_audio.mp3
└── README.md                 # Project README (this file)

Author

Ken Popkin
GitHub: @krpopkin
Email: krpopkin@gmail.com