audio_to_text_tool.py

import base64
import tempfile
from openai import OpenAI
from langchain.tools import tool
from constants import OPENAI_KEY
import tempfile
import os
import openai
from openai import OpenAI
from langchain.tools import tool
import yt_dlp
from utils import get_bytes, get_text_file_contents, get_base64

# Initialize OpenAI client (uses OPENAI_API_KEY from environment or explicitly)
client = OpenAI(api_key=OPENAI_KEY)
 
@tool
def audio_to_text(base64_audio_path: str) -> str:
    """
    Transcribes an audio file (base64-encoded text stored in a file) using OpenAI's Whisper API.

    Args:
        base64_audio_path (str): Path to a file containing base64-encoded audio as text.

    Returns:
        str: The transcribed text.
    """
    try:
        # Read base64 string
        with open(base64_audio_path, "r") as f:
            base64_str = f.read()

        # Decode base64 to bytes
        audio_bytes = base64.b64decode(base64_str)

        # Save audio bytes to temp file (must be supported format: mp3, m4a, wav, etc.)
        with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as temp_audio:
            temp_audio.write(audio_bytes)
            temp_audio_path = temp_audio.name

        # Transcribe using OpenAI Whisper API
        with open(temp_audio_path, "rb") as audio_file:
            transcript = client.audio.transcriptions.create(
                model="whisper-1",
                file=audio_file,
                response_format="text"
            )

        return transcript.strip()

    except Exception as e:
        return f"An error occurred during transcription: {str(e)}"

@tool
def audio_to_text_from_youtube(youtube_url: str) -> str:
    """
    Downloads audio from a YouTube video and transcribes it using OpenAI Whisper API.

    Args:
        youtube_url (str): URL of the YouTube video.

    Returns:
        str: Transcribed text.
    """
    try:
        with tempfile.TemporaryDirectory() as tmpdir:
            audio_stem = os.path.join(tmpdir, "audio")
            audio_output_path = audio_stem + ".mp3"

            BASE_DIR = os.path.dirname(os.path.abspath(__file__))
            cookies_path = os.path.join(BASE_DIR, 'files', 'cookies.txt')

            print('cookies: \n' + get_text_file_contents(cookies_path))

            ydl_opts = {
                "format": "bestaudio/best",
                "outtmpl": audio_stem,
                "quiet": True,
                "cookiefile": cookies_path,
                "postprocessors": [{
                    "key": "FFmpegExtractAudio",
                    "preferredcodec": "mp3",
                    "preferredquality": "192",
                }],
            }

            with yt_dlp.YoutubeDL(ydl_opts) as ydl:
                ydl.download([youtube_url])

            # Ensure file exists
            if not os.path.exists(audio_output_path):
                raise FileNotFoundError(f"Audio file not created: {audio_output_path}")

            # Transcribe with OpenAI Whisper
            with open(audio_output_path, "rb") as audio_file:
                transcript = client.audio.transcriptions.create(
                    model="whisper-1",
                    file=audio_file,
                    response_format="text"
                )

            return transcript.strip()

    except Exception as e:
        return f"An error occurred during YouTube transcription: {str(e)}"

if __name__ == "__main__":
    # Example: path to a text file that contains base64-encoded audio (e.g., base64_audio.txt)
    base64_audio_file_path = r"C:\tmp\ibm\99c9cc74-fdc8-46c6-8f8d-3ce2d3bfeea3.mp3.b64"

    # Call the tool function
    transcription = audio_to_text(base64_audio_file_path)

    # Print the result
    print("Transcription result:")
    print(transcription)