README

Generate speech to text transcription using OpenAI's general purpose speech recognition model Whisper.

Description

From command line, specify the input, working, and output directories. Whisper requires the input files to be constrained to a maximum file size. This utility will pre-process the files in the input directory to split the files into multiple parts if they exceed the maximum input size restriction.

Installation

1. Requires Python version 3.12 or previous

2. Install python-ffmpeg

pip3 install python-ffmpeg

3. Install pytorch with CUDA support (see pytorch documentation).

4. Install whisper

pip3 install git+https://github.com/openai/whisper.git

Usage

Specify the import directory with MP3 files. Specify the working directory where files are either copied (less than 25 MB) or segmented (greater than 25 MB). Specifying the directory where the output text files are saved. The trext files contain the dict output of the Whisper transcription call.

python main.py -i "./test-data/input" -t "./test-data/working" -o "./test-data/output" -v