src/speech_to_text.py

from transformers import pipeline
import librosa # Or soundfile
import os

# Initialize the ASR pipeline with a specific model
# Using a smaller Whisper model for quicker setup, but larger models offer better accuracy
asr_pipeline = pipeline(
    "automatic-speech-recognition",
    model="openai/whisper-tiny.en",
)

def transcribe_audio(audio_filepath):
    """
    Transcribes an audio file using the Hugging Face ASR pipeline.
    """
    try:
        transcription = asr_pipeline(audio_filepath, return_timestamps=True)
        return transcription["text"]
    except Exception as e:
        return f"Error during transcription: {e}"

# Example usage:
if __name__ == "__main__":
    audio_file = "./downloaded_files/1f975693-876d-457b-a649-393859e79bf3.mp3"

    if os.path.exists(audio_file): # Check if the (placeholder or real) file exists
        print(f"Attempting to transcribe: {audio_file}")
        transcribed_text = transcribe_audio(audio_file)
        print(f"Transcription:\n{transcribed_text}")
    else:
        print(f"File not found: {audio_file}. Please provide a valid audio file.")