handler.py

# handler.py

from typing import Dict, Any
from transformers import pipeline

class EndpointHandler:
    def __init__(self, model_path: str = ""):
        """
        Load the MMS-TTS pipeline once at startup.
        transformers>=4.33.0 is required for MMS-TTS support.
        """
        self.tts = pipeline(
            task="text-to-speech",
            model=model_path,
            device=0,
            # device_map="auto"  # optional: to leverage GPU if available
        )

    def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]:
        """
        data: {"inputs": "<text to synthesize>"}
        Returns: {"wav": <binary audio>, "sampling_rate": <int>}
        """
        text = data.get("inputs", "")
        # Run TTS; returns a dict with "wav" and "sampling_rate"
        result = self.tts(text)
        audio = result["audio"]
        return {
            "array": audio.T.tolist(),  # transpose if needed to fix ushort format:contentReference[oaicite:6]{index=6}
            "sampling_rate": result["sampling_rate"],
        }