Audio Transcriber

A TypeScript application that transcribes audio files using OpenAI's Whisper API. It supports splitting large audio files into chunks to comply with the 25MB API limit.

Features

• Transcribes WAV and MP3 audio files using OpenAI's Whisper API
• Automatically splits large audio files into chunks
• Adds timestamps to the transcription
• Supports various transcription options (language, prompt, temperature)
• Easy command-line interface

Prerequisites

• Node.js (v14 or higher)
• FFmpeg installed on your system
• OpenAI API key

Installation

1. Clone this repository:

   git clone https://github.com/gtofig/whisper-transcribe.git
   cd whisper-transcribe

2. Install dependencies:

   npm install

3. Create a .env file in the root directory with your OpenAI API key:

   OPENAI_API_KEY=your_openai_api_key_here
   

4. Build the project:

   npm run build

Usage

Basic Usage

npm start -- -i /path/to/your/audio/file.mp3

The transcription will be saved to ./transcriptions/file_transcription.txt.

Advanced Options

npm start -- \
  -i /path/to/your/audio/file.mp3 \
  -o /path/to/output/directory \
  -m 20 \
  -l en \
  -p "This is a discussion about technology." \
  -t 0.2

Options

• -i, --input: Path to the input audio file (required)
• -o, --output: Output directory for transcriptions (default: ./transcriptions)
• -m, --maxChunkSize: Maximum chunk size in MB (default: 25)
• -l, --language: Language of the audio (ISO-639-1 code)
• -p, --prompt: Prompt to guide the transcription
• -t, --temperature: Temperature for the OpenAI API (default: 0)

Development

For development with hot-reloading:

npm run dev -- -i /path/to/your/audio/file.mp3

Dependencies

• openai: Official OpenAI API client
• fluent-ffmpeg: Node.js wrapper for FFmpeg
• fs-extra: Enhanced file system operations
• dotenv: Environment variable management
• yargs: Command-line argument parsing

License

MIT