simple faster-whisper API server for GCP Cloud Run GPU

server.py

from flask import Flask, request, jsonify, Response
from faster_whisper import WhisperModel
import subprocess

app = Flask(__name__)

# Initialize the faster-whisper model
model = WhisperModel("small", device="cuda")  # Use 'cuda' to leverage the GPU

@app.route('/transcribe', methods=['POST'])
def transcribe():
    file = request.files['audio']

    # Set task to 'translate' for translation to English
    segments, info = model.transcribe(file, task="translate", language=None)

    # Return translated transcriptions
    return jsonify({
        "duration": info.duration,
        "translation": [segment.text for segment in segments]
    })

@app.route('/stats', methods=['GET'])
def gpu_stats():
    # Run the nvidia-smi command
    result = subprocess.run(['nvidia-smi'], stdout=subprocess.PIPE)
    gpu_info = result.stdout.decode('utf-8')

    # Return the output of nvidia-smi as plain text
    return Response(gpu_info, mimetype='text/plain')

if __name__ == '__main__':
    app.run(host='0.0.0.0', port=8080)