Audio-Transcriber

Inspired by https://github.com/Black-coffe/AudioExtractor-and-Transcriber

This repository contains a comprehensive script for handling video to transcribe it into text. You can do several task by them:

1. Get transcription from video file (mp42text)
2. Extract audio from video file (mp42wav)
3. Get text transcription from audio file (wav2text)

Features

• Extract Audio from Video: Utilizes moviepy to extract audio tracks from video files.
• Process Audio Track: Leverages pydub for audio processing, including conversion to mono and setting the sampling rate to 16kHz.
• Transcribe Audio to Text: Employs vosk, a powerful tool for speech recognition, to convert audio into text.
• Text Processing: Saves the transcribed text and can split it into manageable parts for easier handling.

Use Cases

This script is ideal for:

• Creating text versions of video lectures and podcasts.
• Automating subtitle creation for videos.
• Extracting meaningful quotes from video content.
• Preparing data for analysis in machine learning projects.

Technical Requirements

Libraries and Models

• moviepy: For video file manipulation.
• pydub: For audio processing tasks. Requires FFmpeg for operations like format conversion.
• vosk: For accurate speech recognition. You will need to download a model compatible with your audio content's language from Vosk Models.

FFmpeg

FFmpeg is a complete, cross-platform solution to record, convert and stream audio and video. It's required for pydub to process audio files. You can download FFmpeg builds from here.

Vosk Models

Vosk offers a variety of models that support different languages and are optimized for various audio types. You can download the appropriate model for your project from Vosk Models. Unpack the model into the models/ folder and pass its path with -m, e.g. -m models/vosk-model-ru-0.42. Save the model to model folders.

Installation

Clone this repository and install the required dependencies:

git clone https://github.com/gs1571/audio-transcriber.git
cd audio-transcriber
pip install -r requirements.txt

Additionally, you need to install FFmpeg. On macOS, you can install FFmpeg using Homebrew:

brew install ffmpeg

Usage

Everything runs through one command-line entry point, at.py. See python at.py <command> --help for all options.

Command	What it does
mp42text	Video → text in one step (extract audio, convert, transcribe)
mp42wav	Extract audio from a video as 16 kHz mono WAV, ready for recognition
mp42mp3	Extract audio from a video as MP3
wav2text	Transcribe a WAV file
text2parts	Split a transcription into parts of a fixed size (4000 characters by default)

# video straight to text; the result lands next to the video as lecture.txt
python at.py mp42text -s lecture.mp4 -m models/vosk-model-ru-0.42

# keep the raw Vosk JSON too (word timings and confidence)
python at.py mp42text -s lecture.mp4 -m models/vosk-model-ru-0.42 -dj lecture.json

# split the transcription into lecture_part_1.txt, lecture_part_2.txt, ...
python at.py text2parts -s lecture.txt

Credits

The command-line interface and the atlib package were contributed by Grigorii Solovev (@gs1571).