audio_transcriber.py

import os
import requests
from smolagents import Tool

class AudioTranscriptionTool(Tool):
    name = "audio_transcriber"
    description = "Transcribe a given audio file in mp3 or wav format to text using Whisper via Hugging Face API."
    inputs = {
        "file_path": {
            "type": "string",
            "description": "Path to the audio file (must be .mp3 or .wav)"
        }
    }
    output_type = "string"

    def __init__(self):
        super().__init__()
        self.api_url = "https://api-inference.huggingface.co/models/openai/whisper-large"
        self.headers = {
            "Authorization": f"Bearer {os.getenv('HF_API_TOKEN')}"
        }

    def forward(self, file_path: str) -> str:
        try:
            with open(file_path, "rb") as audio_file:
                audio_bytes = audio_file.read()

            response = requests.post(
                self.api_url,
                headers=self.headers,
                data=audio_bytes,
                timeout=60
            )
            if response.status_code == 200:
                result = response.json()
                # The exact key depends on the model; usually 'text' for whisper
                transcription = result.get("text", None)
                if transcription:
                    return transcription.strip()
                else:
                    return "Error: No transcription found in the response."
            else:
                return f"Error transcribing audio: {response.status_code} {response.text}"
        except Exception as e:
            return f"Error transcribing audio: {e}"