Whisper API

A FastAPI-based web service that provides speech-to-text transcription and translation capabilities using OpenAI's Whisper model. This service offers RESTful endpoints for converting audio files into text and translating non-English audio to English text.

Features

• Audio Transcription: Convert speech in audio files to text in the original language
• Audio Translation: Translate non-English audio to English text
• Multi-format Support: Support for various audio and video formats
• OpenAI-Compatible API: RESTful endpoints compatible with OpenAI's audio API
• Docker Support: Containerized deployment for easy scaling
• Robust Error Handling: Comprehensive input validation and error responses

Supported File Formats

• Audio: .mp3, .wav, .mpga, .webm, .m4a
• Video: .mp4, .mpeg
• Size Limit: Configurable file size limits for optimal performance

Prerequisites

• Python 3.11 or higher
• FFmpeg (for audio processing)
• Docker (optional, for containerized deployment)

Quick Start

Local Installation

1. Clone the repository:

git clone https://github.com/duytechie/whisper-api-server.git
cd whisper-api

2. Install dependencies:

# Install CPU-optimized PyTorch
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu

# Install Whisper and FastAPI
pip install -U openai-whisper
pip install "fastapi[standard]"

3. Install FFmpeg:

Ubuntu/Debian:

sudo apt update && sudo apt install ffmpeg

macOS:

brew install ffmpeg

4. Run the application:

python main.py

The application will be available at http://localhost:8000

Docker Deployment

# Build the Docker image
docker build -t whisper-api .

# Run the container
docker run -d --name whisper-service -p 8000:8000 whisper-api

API Usage

Interactive Documentation

Access the automatically generated API documentation at:

• Swagger UI: http://localhost:8000/docs
• ReDoc: http://localhost:8000/redoc

API Endpoints

Health Check

GET /

Transcription

Convert audio to text in the original language:

curl -X POST http://localhost:8000/v1/audio/transcriptions \
  -F "file=@example.mp3" \
  -F "model=small"

Translation

Translate audio to English text:

curl -X POST http://localhost:8000/v1/audio/translations \
  -F "file=@spanish_audio.mp3" \
  -F "model=small"

Available Whisper Models

Model	Parameters	Size	Relative Speed
tiny	39 M	~1 GB	~32x
base	74 M	~1 GB	~16x
small	244 M	~2 GB	~6x
medium	769 M	~5 GB	~2x
large	1550 M	~10 GB	1x

Development

Local Development with Auto-reload

uvicorn whisper_api:app --reload --host 0.0.0.0 --port 8000

Technology Stack

• FastAPI: Modern, fast web framework for building APIs
• OpenAI Whisper: Automatic speech recognition system
• PyTorch: Machine learning framework (CPU-optimized)
• Uvicorn: ASGI server for FastAPI applications
• Docker: Containerization platform
• FFmpeg: Audio/video processing library