stt_handler.pyc

# rentbot/stt_handler.py
import whisper
import numpy as np
import asyncio
import os
from io import BytesIO

# Load the model once when the module is imported
print("Loading Whisper model...")
model = whisper.load_model("base.en")
print("Whisper model loaded.")

async def transcribe_audio_chunk(audio_chunk: np.ndarray) -> str:
    """
    Transcribes an audio chunk using Whisper.
    Runs the blocking whisper call in a separate thread.
    """
    # The audio data is 16-bit PCM, 8000 Hz. Whisper expects float32.
    audio_float32 = audio_chunk.astype(np.float32) / 32768.0

    # Using an in-memory buffer
    wav_buffer = BytesIO()
    
    # We must provide the sample rate to whisper's transcribe function
    loop = asyncio.get_event_loop()
    result = await loop.run_in_executor(
        None,  # Use the default executor (a ThreadPoolExecutor)
        lambda: model.transcribe(
            audio_float32, 
            language="en", 
            fp16=False # Set to False if not using a GPU
        )
    )
    
    return result.get("text", "").strip()