FastAPI WhisperX Transcription Service

This repository provides a FastAPI-based API for audio transcription, alignment, and speaker diarization using WhisperX.

👉 Try the live demo on Hugging Face Spaces

---

📁 Project Structure

fastapi/
├── app/
│   └── main.py                # FastAPI app with /transcribe endpoint
├── test_audios/               # Example audio files for testing
│   ├── BernhardtCrescent.wav
│   ├── BlackStone_en_in.mp4
│   ├── BlackStone_en_in.wav
│   ├── fillicafe.wav
│   └── harvard.wav
├── requirements.txt           # Python dependencies
├── dockerfile                 # Docker setup

---

⚙️ Setup & Local Development

1. Clone the Repository

git clone <your-repo-url>
cd fastapi

2. Create and Activate a Virtual Environment

Option A: Python

python3 -m venv .venv
source .venv/bin/activate

Option B: Conda (Recommended)

Build and run the container:

conda create --name whisperx_api python==3.10 
conda activate whisperx_api

---

3. Install Python Dependencies

pip install --upgrade pip
pip install -r requirements.txt

4. Install System Dependencies

• ffmpeg is required for audio processing.
• On Ubuntu/Debian:

  sudo apt-get update && sudo apt-get install -y ffmpeg git

---

🚀 Running the API

Option A: Local (Recommended for development)

uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload

Option B: Docker

Build and run the container:

docker build -t curify_fastapi .  
docker run -p 8000:8000 curify_fastapi

---

🧪 Testing the API

1. Using curl

Upload an audio file from test_audios/ for transcription:

curl -X POST "http://localhost:8000/transcribe" \
  -H "Content-Type: multipart/form-data" \
  -F "file=@test_audios/harvard.wav"

2. Using Swagger UI

Visit http://localhost:8000/docs for interactive API documentation.

---

📝 Endpoint Overview

• POST /transcribe
  Upload an audio file and receive a transcript with speaker labels and timestamps.

---

To extract audio from video using ffmpeg

pip install ffmpeg
ffmpeg -i BlackStone_en_in.mp4 -ar 16000 -ac 1 BlackStone_en_in.wav

🧹 Cleanup

Temporary files are automatically deleted after each request.

---

🛠️ Notes

• The WhisperX model is loaded once at startup for efficiency.
• Diarization uses a Hugging Face token (edit in main.py if needed).
• For best results, use clear audio files (see test_audios/ for examples).

---