WhisperX ASR with FastAPI

WhisperX ASR is a production-ready automatic speech recognition (ASR) app powered by WhisperX and FastAPI. It provides a web UI and REST API for transcribing audio files with word-level timestamps, multi-language support, and GPU acceleration.

For advanced configuration, model options, and setup instructions, see the original WhisperX repository: https://github.com/m-bain/whisperX

---

🚀 Features

• FastAPI backend for robust, scalable API serving
• WhisperX model for accurate speech-to-text transcription
• Web UI for uploading audio, recording, and viewing results
• Supports multiple audio formats: WAV, MP3, M4A, FLAC, OGG, WEBM
• Language selection (auto-detect or manual)
• Batch size control for performance tuning
• Word-level timestamps and segments in results
• GPU acceleration (if available)
• Health check endpoint for monitoring model status

---

🛠️ Technologies

• FastAPI
• WhisperX
• Uvicorn
• Python 3.11.5
• HTML/CSS/JS for the frontend

---

📦 Installation & Usage

Requirements

• Python 3.11.5+
• uvicorn
• uv package manager
• CUDA-enabled GPU for acceleration (you can run on CPU but slower)

Setup

# Clone the repository
git clone https://github.com/romanyn36/whisperx-automatic-speech-recognition.git
cd whisperx-automatic-speech-recognition

# setup virtual environment
uv sync 
.\.venv\Scripts\activate # Windows
source .venv/bin/activate  # Linux/Mac

# copy the example env file and modify as needed
cp .env.example .env

# Start the server
uv run python main.py --reload

# access the web UI at http://localhost:8000/static/

---

📝 API Endpoints

Health Check

GET /health

• Returns model status, GPU info, and readiness.

Supported Languages

GET /languages

• Lists available transcription languages.

Transcribe Audio

POST /transcribe

• Parameters:
  • file: Audio file (WAV, MP3, M4A, FLAC, OGG, WEBM)
  • language: Language code (e.g., en, auto)
  • batch_size: Integer (1–32)
• Returns:
  • transcription: Full text
  • language: Detected language
  • processing_time: Seconds
  • segments: List of segments with timestamps
  • word_segments: List of word-level timestamps

---

🌐 Web UI

Open http://localhost:8000/ in your browser. Features:

• Upload or record audio
• Select language and batch size
• View transcription, segments, and word-level timestamps

---

⚙️ Configuration

You can adjust model size, alignment, and file size limits via environment variables in .env:

• MODEL_SIZE (e.g., large-v2, medium, tiny) in my gtx 1650 GPU the large-v3-turbo works excellent
• ENABLE_ALIGNMENT (true/false)
• MAX_FILE_SIZE_MB (default: 100)
• API_KEY (optional, disabled by default)

---

What's Next

as this project is still ongoing and this is just the first development phase there are many features to come

• currently working deploy on gpu (runpod)
• add production ready docker support
• built github actions for auto deployment
• testing and improvements among others whisperx features

📄 License

This project is licensed under the MIT License. See LICENSE.

---

👤 Author & Contact

Made by Romani Nasrat – AI Engineer & Backend Developer
romani.nasrat@gmail.com | Portfolio