Function run_tts

cli/inference.py:64–107

Perform TTS inference and save the generated audio.

(args)

def run_tts(args):
    """Perform TTS inference and save the generated audio."""
    logging.info(f"Using model from: {args.model_dir}")
    logging.info(f"Saving audio to: {args.save_dir}")

    # Ensure the save directory exists
    os.makedirs(args.save_dir, exist_ok=True)

    # Convert device argument to torch.device
    if platform.system() == "Darwin" and torch.backends.mps.is_available():
        # macOS with MPS support (Apple Silicon)
        device = torch.device(f"mps:{args.device}")
        logging.info(f"Using MPS device: {device}")
    elif torch.cuda.is_available():
        # System with CUDA support
        device = torch.device(f"cuda:{args.device}")
        logging.info(f"Using CUDA device: {device}")
    else:
        # Fall back to CPU
        device = torch.device("cpu")
        logging.info("GPU acceleration not available, using CPU")

    # Initialize the model
    model = SparkTTS(args.model_dir, device)

    # Generate unique filename using timestamp
    timestamp = datetime.now().strftime("%Y%m%d%H%M%S")
    save_path = os.path.join(args.save_dir, f"{timestamp}.wav")

    logging.info("Starting inference...")

    # Perform inference and save the output audio
    with torch.no_grad():
        wav = model.inference(
            args.text,
            args.prompt_speech_path,
            prompt_text=args.prompt_text,
            gender=args.gender,
            pitch=args.pitch,
            speed=args.speed,
        )
        sf.write(save_path, wav, samplerate=16000)

    logging.info(f"Audio saved at: {save_path}")


if __name__ == "__main__":