from typing import Dict
from faster_whisper import WhisperModel

class EndpointHandler:
    def __init__(self, model_dir=None):
        # If a model_dir is provided, use it; otherwise, default to 'large-v2'.
        model_size = "large-v2" if model_dir is None else model_dir
        self.model = WhisperModel(model_size, device="cpu")

    def __call__(self, data: Dict) -> Dict[str, str]:
        # Process the input data expected to be in 'inputs' key containing audio file bytes
        audio_bytes = data["inputs"]
        
        # Perform transcription using the model
        segments, info = self.model.transcribe(audio_bytes)
        
        # Compile the results into a text string and extract language information
        text = " ".join(segment.text for segment in segments)
        language_code = info.language
        language_prob = info.language_probability

        # Compile the response dictionary
        result = {
            "text": text,
            "language": language_code,
            "language_probability": language_prob
        }
        
        return result