models/audio_gen.py

import torch
from transformers import VitsModel, AutoTokenizer
import numpy as np

class AudioGenerator:
    def __init__(self):
        print("Initializing Audio Generator...")
        self.device = "cuda" if torch.cuda.is_available() else "cpu"
        print(f"Using device: {self.device}")
        
        # Load model and tokenizer
        self.model_name = "facebook/mms-tts-eng"
        print(f"Loading model {self.model_name}...")
        
        self.tokenizer = AutoTokenizer.from_pretrained(self.model_name)
        self.model = VitsModel.from_pretrained(self.model_name).to(self.device)
        print(f"Model loaded and moved to {self.device}")

    def generate_audio(self, text, voice_preset=None):
        """
        Generate audio from text using the MMS-TTS model
        
        Args:
            text (str): The text to convert to speech
            voice_preset (str): Not used in this implementation
            
        Returns:
            tuple: (audio_array, sample_rate)
        """
        try:
            print(f"Generating audio on {self.device}...")
            
            # Tokenize the input text
            inputs = self.tokenizer(text, return_tensors="pt").to(self.device)
            
            with torch.no_grad():
                output = self.model(**inputs).waveform
            
            # Convert to numpy array and normalize
            audio = output.cpu().numpy().squeeze()
            audio = (audio * 32767).astype(np.int16)
            
            return audio, self.model.config.sampling_rate

        except Exception as e:
            return f"Error generating audio: {str(e)}"