FreeScribe Speech-to-Text Server

This software is released under the AGPL-3.0 license
Copyright (c) 2023-2024 Braedon Hendy

Part of the ClinicianFOCUS initiative, a collaboration with Conestoga College Institute of Applied Learning and Technology's CNERG+ applied research project.

Overview

FreeScribe is an open-source speech-to-text server that utilizes OpenAI's Whisper model for audio transcription. It provides a robust API endpoint for converting audio files to text, with support for various audio formats including MP3, WAV, and WebM.

Features

• Fast audio transcription using Whisper AI
• Support for multiple audio formats (MP3, WAV, WebM)
• Rate limiting protection
• API key authentication
• GPU support for faster processing
• Docker containerization
• Reverse proxy setup with Caddy

Prerequisites

• Docker and Docker Compose
• NVIDIA GPU (optional, for GPU acceleration)
• NVIDIA Container Toolkit (if using GPU)

Configuration

Environment Variables

Create a .env file with the following variables:

WHISPER_MODEL=medium    # Whisper model size (tiny, base, small, medium, large) (Default: medium)
WHISPER_PORT=2224      # Port for the service (Default: 2224)
WHISPER_HOST=0.0.0.0   # Host address (Default: 0.0.0.0)
UVICORN_WORKERS=1      # Number of Uvicorn workers (Default: 1)
DEBUG_MODE=False       # Debug mode toggle switch. Set to true to enable debug mode. (Default: False)

Deployment Options

CPU-Only Deployment

docker-compose -f docker-compose.cpu.yml up --build

GPU-Enabled Deployment

docker-compose up --build

Container Structure

The application consists of two main containers:

1. speech-container

   • Runs the FastAPI application
   • Handles audio transcription
   • Configurable for CPU or GPU usage

2. caddy

   • Reverse proxy
   • Handles HTTPS termination
   • Manages incoming traffic

API Usage

Endpoint: POST /health

API to check if container is running.

Response:

{
  "status": "ok"
}

Endpoint: POST /whisperaudio

Transcribe an audio file to text.

Request:

• Method: POST
• Content-Type: multipart/form-data
• Authorization: Bearer <api_key>
• Body: audio file (MP3, WAV, or WebM)

Response:

{
  "text": "Transcribed text from the audio file."
}

Rate Limiting

• Default rate limit: 1 request per second
• Rate limit errors return 429 status code

License

This project is licensed under the AGPL-3.0 License - see the LICENSE file for details.