import os
from smolagents import tool  # Assuming smolagents.tool is the correct decorator


@tool
def transcribe_mp3_audio_file(mp3_file_path: str) -> str:
    """
    Transcribes an MP3 audio file using OpenAI's Whisper 'base' model.

    Args:
        mp3_file_path (str): The absolute local path to the MP3 audio file.
                             This path should be obtained from the 'File Information' section 
                             if the file was downloaded by the agent.

    Returns:
        str: The transcribed text from the audio file, or an error message if transcription fails.
    """
    try:
        import whisper  # Attempt to import whisper only when the tool is called
    except ImportError:
        return ("Error: The 'openai-whisper' library is required but not installed. "
                "Please install it using 'pip install openai-whisper' and ensure ffmpeg is also installed.")

    if not os.path.exists(mp3_file_path):
        return f"Error: Audio file not found at the specified path: '{mp3_file_path}'. Please verify the path."

    if not mp3_file_path.lower().endswith(".mp3"):
        return f"Error: The provided file path '{mp3_file_path}' does not appear to be an MP3 file. This tool currently only supports .mp3 files."

    try:
        print(
            f"AudioTool: Loading Whisper 'base' model to transcribe '{mp3_file_path}'...")
        # You can choose different model sizes: "tiny", "base", "small", "medium", "large"
        # "base" is a good balance of speed and accuracy for many use cases.
        # Larger models are more accurate but slower and require more resources.
        model = whisper.load_model("base")

        print(f"AudioTool: Transcribing audio from '{mp3_file_path}'...")
        # fp16=False can improve compatibility on CPU
        result = model.transcribe(mp3_file_path, fp16=False)

        transcribed_text = result.get("text", "")
        if transcribed_text:
            print(
                f"AudioTool: Transcription successful for '{mp3_file_path}'.")
            return transcribed_text
        else:
            # This case might occur if the audio is silent or whisper couldn't detect speech
            return f"Notice: Transcription resulted in empty text for '{mp3_file_path}'. The audio might be silent or contain no clear speech."

    except FileNotFoundError:  # Should be caught by os.path.exists, but as a fallback for whisper's internal handling
        return f"Error: Whisper could not find the audio file at path: '{mp3_file_path}' (even if it was initially detected)."
    except Exception as e:
        # Check if the error is due to ffmpeg not being found
        if "ffmpeg" in str(e).lower() and ("not found" in str(e).lower() or "not installed" in str(e).lower()):
            return ("Error during transcription: ffmpeg not found. "
                    "OpenAI Whisper requires ffmpeg to be installed and in your system's PATH. "
                    f"Details: {type(e).__name__} - {str(e)}")
        return f"Error during audio transcription for '{mp3_file_path}': {type(e).__name__} - {str(e)}"


if __name__ == '__main__':
    # This is a placeholder for testing.
    # To test, you would need an actual MP3 file.
    # For example:
    # test_mp3_path = "path/to/your/test_audio.mp3"
    # if os.path.exists(test_mp3_path):
    #     print(f"--- Testing with MP3 file: {test_mp3_path} ---")
    #     transcript = transcribe_mp3_audio_file(test_mp3_path)
    #     print("Transcription Result:")
    #     print(transcript)
    # else:
    #     print(f"Test MP3 file not found at: {test_mp3_path}. Cannot run local test.")

    print("Audio transcription tool defined. To test, provide a path to an MP3 file in the example block.")
    # Test with a non-existent file
    print("\n--- Testing with non-existent file ---")
    print(transcribe_mp3_audio_file("non_existent_file.mp3"))
    # Test with a non-mp3 file
    print("\n--- Testing with non-MP3 file extension ---")
    # Create a dummy file for this test
    dummy_file = "dummy.txt"
    with open(dummy_file, "w") as f:
        f.write("this is not an mp3")
    print(transcribe_mp3_audio_file(dummy_file))
    os.remove(dummy_file)