Add return_timestamps=True to whisper pipeline

app.py

@@ -29,10 +29,18 @@ def transcribe_audio(audio_file):
         if audio.shape[0] > 1:  # Check if multi-channel
             audio = torch.mean(audio, dim=0, keepdim=True) # Average channels
 
-        # Transcribe the audio
-        transcription = whisper_pipeline(audio.squeeze().numpy())["text"] # .squeeze() removes extra dimensions
-
-        return transcription
+        # Long-Form Transcription with Timestamps
+        transcription = whisper_pipeline(audio.squeeze().numpy(), return_timestamps=True)
+
+        # Format the output with timestamps (Improved)
+        formatted_transcription = ""
+        for segment in transcription["chunks"]:
+            start = segment["timestamp"][0]
+            end = segment["timestamp"][1]
+            text = segment["text"]
+            formatted_transcription += f"[{start:.2f} - {end:.2f}] {text}\n"  # Nicer formatting
+
+        return formatted_transcription
 
     except Exception as e:
         return f"An error occurred: {e}"