Update app.py

app.py

@@ -327,67 +327,43 @@ class AnalyzeImageTool(Tool):
             return response
         except Exception as e:
             return f"Vision error: {e}"
-# ─────────────────────────────────────────────────────────────
-# Tool – Audio transcription (MP3/WAV → text via Gemini)
-# ─────────────────────────────────────────────────────────────
+
 class TranscribeAudioTool(Tool):
     name = "transcribe_audio"
     description = (
-        "Download an audio file (MP3/WAV/M4A) and transcribe its speech to text. "
-        "Use for questions that reference attached audio recordings or voice memos."
+        "Convert speech from audio files into text. "
+        "Use for podcasts, interviews, voice memos, or any recorded speech content."
     )
     inputs = {
         "source": {
             "type": "string",
-            "description": "URL or local file path of the audio file",
+            "description": "Audio file URL or local path (MP3/WAV/M4A/FLAC/OGG)",
         },
     }
     output_type = "string"
 
+    def __init__(self):
+        from huggingface_hub import InferenceClient
+        self.client = InferenceClient(
+            model="openai/whisper-small",
+            token=os.getenv("HF_TOKEN")
+        )
+
     def forward(self, source: str) -> str:
         try:
             if source.startswith("http"):
                 r = requests.get(source, timeout=30)
                 r.raise_for_status()
-                audio_bytes = r.content
-                ext = source.split("?")[0].rsplit(".", 1)[-1].lower() or "mp3"
+                audio_data = r.content
             else:
                 with open(source, "rb") as f:
-                    audio_bytes = f.read()
-                ext = source.rsplit(".", 1)[-1].lower()
-        except Exception as e:
-            return f"Could not load audio: {e}"
-
-        b64 = base64.b64encode(audio_bytes).decode()
-        api_key = os.getenv("GEMINI_API_KEY") or os.getenv("API_KEY", "")
-        payload = {
-            "model": MODEL_NAME,
-            "max_tokens": 2048,
-            "messages": [{
-                "role": "user",
-                "content": [
-                    {"type": "input_audio",
-                     "input_audio": {"data": b64, "format": ext}},
-                    {"type": "text",
-                     "text": (
-                         "Please transcribe this audio recording completely and accurately. "
-                         "Include every word spoken. Return only the transcription, nothing else."
-                     )},
-                ],
-            }],
-        }
-        try:
-            resp = requests.post(
-                f"{API_BASE}chat/completions",
-                headers={"Authorization": f"Bearer {api_key}",
-                         "Content-Type": "application/json"},
-                json=payload, timeout=90,
-            )
-            if resp.status_code == 200:
-                return resp.json()["choices"][0]["message"]["content"].strip()
-            return f"Audio API error: HTTP {resp.status_code} {resp.text[:300]}"
+                    audio_data = f.read()
+            
+            response = self.client.automatic_speech_recognition(audio_data)
+            return response.get("text", "No speech detected")
         except Exception as e:
-            return f"Could not transcribe audio: {e}"
+            return f"Transcription error: {e}"
+            
 
 
 # ─────────────────────────────────────────────────────────────
@@ -533,7 +509,7 @@ class BasicAgent:
                 YouTubeTranscriptTool(),
                 SafeVisitWebpageTool(),
                 AnalyzeImageTool(),
-                # TranscribeAudioTool(),
+                TranscribeAudioTool(),
                 ReadExcelFileTool(),
                 ReadTextFileTool(),
                 WikipediaTablesTool(),