tts.py

import torch
from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
import logging
import numpy as np
import soundfile as sf
from huggingface_hub import hf_hub_download

# Set up logging
logging.basicConfig(level=logging.DEBUG)

MODEL_ID = "facebook/mms-tts-fao"

# Try to load the model and processor
try:
    processor = Wav2Vec2Processor.from_pretrained(MODEL_ID)
    model = Wav2Vec2ForCTC.from_pretrained(MODEL_ID)
    logging.info("Model and processor loaded successfully.")
except Exception as e:
    logging.error(f"Error loading model or processor: {e}")
    raise

def synthesize_speech(text):
    try:
        # Ensure text is not empty
        if not text.strip():
            logging.error("Text input is empty.")
            return None

        inputs = processor(text, return_tensors="pt")
        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        model.to(device)
        inputs = inputs.to(device)

        with torch.no_grad():
            speech = model.generate(**inputs)
        
        logging.info("Speech generated successfully.")

        # Decode the generated speech and save to an audio file
        waveform = speech.cpu().numpy().flatten()
        # Normalize waveform to the range [-1, 1]
        waveform = np.clip(waveform, -1.0, 1.0)
        
        # Convert waveform to audio format that Gradio can handle
        audio_path = "output.wav"
        sf.write(audio_path, waveform, 16000)
        return audio_path
    except Exception as e:
        logging.error(f"Error during speech synthesis: {e}")
        return None