| 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.") |