app_srv/audio_processing.py

from transformers import pipeline
import torch
import gc
import librosa

from model_api import clear_gpu_cache, get_device_and_dtype


def transcribe_audio(audio_path: str, device: str = "cuda", torch_dtype: torch.dtype = torch.float16) -> str:
    """
    Transcribes an MP3 audio file.

    Args:
        audio_path: Path to the audio file.
        device: The device to use for transcription (e.g., "cuda" for GPU, "cpu" for CPU).
        torch_dtype: The torch data type to use for model computations.

    Returns:
        The transcribed text.
    """
    try:
        # Create a pipeline with explicit device specification
        pipe = pipeline(
            "automatic-speech-recognition",
            model="openai/whisper-small",
            device=device,
            chunk_length_s=25,
            stride_length_s=2,
            torch_dtype=torch_dtype,
        )

        # Perform transcription
        result = pipe(audio_path, return_timestamps=True)

        # Extract text
        text = result['text']

        return text

    except Exception as e:
        print(f"Error during transcription: {str(e)}")
        return ""

    finally:
        del pipe
        clear_gpu_cache()


if __name__ == "__main__":

    selected_device, selected_dtype = get_device_and_dtype()

    result = transcribe_audio("/workspaces/Video_Analyser/app_srv/downloads/45677153-510d-4f47-95ee-c1b4b0843433/audio.mp3.mp3", selected_device, selected_dtype)
    print(result)