app.py

from flask import Flask, request, jsonify, send_file
from flask_cors import CORS
from transformers import VitsModel, AutoTokenizer
import torch
import scipy.io.wavfile

app = Flask(__name__)
CORS(app)

# --- Load the MMS TTS models and tokenizers ---
# Note: Models are loaded on startup. For a larger number of languages, lazy loading would be better.
MODELS = {
    "ewe": "facebook/mms-tts-ewe",
    "mina": "facebook/mms-tts-gej" # Mina (Gen)
}

loaded_models = {}
loaded_tokenizers = {}

print("Loading models...")
for lang, model_id in MODELS.items():
    print(f"Loading {lang} model: {model_id}")
    try:
        loaded_models[lang] = VitsModel.from_pretrained(model_id)
        loaded_tokenizers[lang] = AutoTokenizer.from_pretrained(model_id)
    except Exception as e:
        print(f"Failed to load {lang} model: {e}")
print("Models loaded successfully.")

@app.route('/tts', methods=['POST'])
def text_to_speech():
    """
    This endpoint receives text and a language code, returning a WAV audio file.
    """
    data = request.get_json()
    text = data.get('text')
    lang = data.get('language', 'ewe')  # Default to Ewe

    if not text:
        return jsonify({"error": "No text provided"}), 400
    
    if lang not in loaded_models:
         return jsonify({"error": f"Language '{lang}' not supported"}), 400

    try:
        model = loaded_models[lang]
        tokenizer = loaded_tokenizers[lang]

        # Tokenize the input text
        inputs = tokenizer(text, return_tensors="pt")

        # Generate the speech waveform
        with torch.no_grad():
            output = model(**inputs).waveform

        # Save the audio file as a WAV file
        audio_file_path = "output.wav"
        sampling_rate = model.config.sampling_rate
        scipy.io.wavfile.write(audio_file_path, rate=sampling_rate, data=output.float().numpy().T)


        return send_file(audio_file_path, mimetype="audio/wav")

    except Exception as e:
        return jsonify({"error": f"An unexpected error occurred: {e}"}), 500

if __name__ == '__main__':
    # It's recommended to use a production-ready WSGI server instead of app.run in production
    app.run(debug=True)