app.py

from flask import Flask, request, jsonify
import requests
import torch
import librosa
from transformers import WhisperProcessor, WhisperForConditionalGeneration

app = Flask(__name__)

# Temporarily using smaller model for faster testing
model_id = "openai/whisper-base"
processor = WhisperProcessor.from_pretrained(model_id)
model = WhisperForConditionalGeneration.from_pretrained(model_id)

device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)

forced_decoder_ids = processor.get_decoder_prompt_ids(language="he", task="transcribe")

def transcribe_audio(audio_url):
    response = requests.get(audio_url)
    with open("temp_audio.wav", "wb") as f:
        f.write(response.content)

    waveform, sr = librosa.load("temp_audio.wav", sr=16000)
    max_duration_sec = 3600
    waveform = waveform[:sr * max_duration_sec]

    chunk_duration_sec = 25
    chunk_size = sr * chunk_duration_sec
    chunks = [waveform[i:i + chunk_size] for i in range(0, len(waveform), chunk_size)]

    partial_text = ""
    for chunk in chunks:
        inputs = processor(chunk, sampling_rate=16000, return_tensors="pt", padding=True)
        input_features = inputs.input_features.to(device)

        with torch.no_grad():
            predicted_ids = model.generate(input_features, forced_decoder_ids=forced_decoder_ids)

        transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
        partial_text += transcription + "\n"

    return partial_text.strip()

@app.route('/transcribe', methods=['POST'])
def transcribe_endpoint():
    data = request.get_json()
    audio_url = data.get('audio_url')
    if not audio_url:
        return jsonify({"error": "Missing 'audio_url' in request"}), 400

    transcription = transcribe_audio(audio_url)

    return jsonify({"transcription": transcription})

if __name__ == '__main__':
    app.run(host="0.0.0.0", port=7860)