feat: add transcribe audio tool

requirements.txt

@@ -5,4 +5,5 @@ smolagents
 duckduckgo-search
 markdownify
 gradio[oauth]
-huggingface_hub
+huggingface_hub
+openai

tools.py

@@ -1,9 +1,14 @@
 import re
 import requests
+import os
+
 from markdownify import markdownify
 from requests.exceptions import RequestException
 from smolagents import tool
 from huggingface_hub import InferenceClient
+from openai import OpenAI
+from urllib.parse import urlparse
+from pathlib import Path
 
 
 @tool
@@ -34,6 +39,7 @@ def visit_webpage(url: str) -> str:
     except Exception as e:
         return f"An unexpected error occurred: {str(e)}"
 
+
 @tool
 def analyze_image(url: str, prompt: str) -> str:
     """Uses a vision model to identify features in an describe an image.
@@ -70,3 +76,67 @@ def analyze_image(url: str, prompt: str) -> str:
         temperature=0.7
     )
     description = response.choices[0].message.content
+
+
+def download_file(url, save_path):
+    """Download a file from a URL and save it locally."""
+    try:
+        response = requests.get(url, stream=True)
+        response.raise_for_status()
+        with open(save_path, "wb") as f:
+            for chunk in response.iter_content(chunk_size=8192):
+                f.write(chunk)
+        return save_path
+    except requests.RequestException as e:
+        raise Exception(f"Failed to download file from {url}: {e}")
+
+
+@tool
+def transcribe_audio(file_path_or_url):
+    """
+    Transcribe an MP3 file using OpenAI Whisper API.
+    Accepts either a local file path or a URL to an MP3 file.
+    """
+    client = OpenAI(
+      api_key=os.environ['OPENAI_API_KEY'],
+    )
+    
+    try:
+        # Check if input is a URL
+        if file_path_or_url.startswith(("http://", "https://")):
+            # Extract filename from URL
+            parsed_url = urlparse(file_path_or_url)
+            filename = os.path.basename(parsed_url.path) or "downloaded_audio.mp3"
+            temp_file_path = os.path.join(os.getcwd(), filename)
+            # Download the file
+            print(f"Downloading file from {file_path_or_url}...")
+            file_path = download_file(file_path_or_url, temp_file_path)
+        else:
+            # Use local file path
+            file_path = file_path_or_url
+            if not os.path.exists(file_path):
+                raise FileNotFoundError(f"Local file {file_path} does not exist.")
+
+        # Check file size (Whisper API limit: 25 MB)
+        file_size = os.path.getsize(file_path) / (1024 * 1024)  # Size in MB
+        if file_size > 25:
+            raise ValueError(f"File size {file_size:.2f} MB exceeds Whisper API limit of 25 MB.")
+
+        # Open and send the file to Whisper API
+        print(f"Transcribing {file_path}...")
+        with open(file_path, "rb") as audio_file:
+            transcription = client.audio.transcriptions.create(
+                model="whisper-1",
+                file=audio_file,
+                response_format="text"
+            )
+
+        # If file was downloaded, clean up
+        if file_path_or_url.startswith(("http://", "https://")):
+            os.remove(file_path)
+            print(f"Cleaned up temporary file: {file_path}")
+
+        return transcription
+
+    except Exception as e:
+        raise Exception(f"Error during transcription: {e}")