Voice Transcription Tool

This tool transcribes or translates audio/video files into text using OpenAI's Whisper model. It supports various languages and can run on both CPU and GPU (CUDA).

Prerequisites

• Python 3.9 or higher
• OpenAI Whisper (openai-whisper)
• PyTorch (torch)

Installation

1. Install Python 3.9 or higher from python.org.

2. Install the required dependencies:

   pip install openai-whisper torch

3. (Optional) For GPU support, ensure you have the appropriate version of PyTorch with CUDA installed. Refer to the PyTorch installation guide.

Usage

Basic Command

python transcript.py \
    --model turbo \
    --voicesource ./source/source.mp4 \
    --lang en \
    --output ./gpt-prompt.txt

Arguments

Argument	Description	Default Value	Required
--lang	Language code of the audio (e.g., en for English, yue for Cantonese).	yue	No
--voicesource	Path to the audio/video file to transcribe.	-	Yes
--output	Path to save the transcription text file. If not provided, the output file will be saved in the same directory as the input file with a .txt extension.	-	No
--model	Whisper model size (e.g., base, small, medium, large, turbo).	base	No
--model_file	Path to a custom Whisper model file.	-	No
--device	Device to use (cuda for GPU or cpu for CPU).	cuda	No

Examples

1. Transcribe an English audio file:

   python transcript.py \
       --voicesource ./audio/english.mp3 \
       --lang en \
       --output ./transcriptions/english.txt

2. Translate a Cantonese audio file to English:

   python transcript.py \
       --voicesource ./audio/cantonese.mp3 \
       --lang yue \
       --translate \
       --output ./translations/cantonese_english.txt

3. Use a custom model:

   python transcript.py \
       --voicesource ./audio/spanish.mp3 \
       --lang es \
       --model_file ./custom_models/spanish_model.pt \
       --output ./transcriptions/spanish.txt

4. Run on CPU:

   python transcript.py \
       --voicesource ./audio/french.mp3 \
       --lang fr \
       --device cpu \
       --output ./transcriptions/french.txt

Output

The tool generates a text file containing the transcription or translation of the input audio/video file. If the --output argument is not provided, the output file will be saved in the same directory as the input file with a .txt extension.

For example:

• Input: ./source/source.mp4
• Output: ./source/source.txt

Supported Models

The following Whisper models are supported:

• tiny
• base
• small
• medium
• large
• turbo

Supported Languages

The tool supports all languages supported by Whisper. Common language codes include:

• English: en
• Cantonese: yue
• Mandarin: zh
• Spanish: es
• French: fr
• German: de

For a full list of supported languages, refer to the Whisper documentation.

Notes

• Ensure your system has sufficient memory (RAM/VRAM) for larger models like large or turbo.
• For GPU usage, ensure CUDA is properly installed and configured.
• If the --translate flag is used, the output will always be in English.

License

This project is licensed under the MIT License. See the LICENSE file for details.