Added number of speakers

app.py

@@ -29,17 +29,24 @@ if torch.cuda.is_available():
 import gradio as gr
 
 
-def transcribe(audio_path):
-    # Run diarization while we wait for Whisper JAX
+# def transcribe(audio_path):
+#     # Run diarization while we wait for Whisper JAX
+#     diarization = diarization_pipeline(audio_path)
+#     #segments = diarization.for_json()["content"]
+#     # Segments = diarization.for_json()["content"]
+#     # Segments = str(diarization)
+#     transcription = "SAML Output"
+#     return diarization
+
+def transcribe(audio_path, num_speakers=2):
+    # Configure the pipeline to use the provided number of speakers
+    diarization_pipeline.n_speakers = num_speakers
+    
+    # Run diarization
     diarization = diarization_pipeline(audio_path)
-    #segments = diarization.for_json()["content"]
-    # Segments = diarization.for_json()["content"]
-    # Segments = str(diarization)
-    transcription = "SAML Output"
+    
     return diarization
 
-
-
 title = "SAML Speaker Diarization ⚡️ "
 
 description = """Combine the speed of Whisper JAX with pyannote speaker diarization to transcribe meetings in super fast time. Demo uses Whisper JAX as an [endpoint](https://twitter.com/sanchitgandhi99/status/1656665496463495168) and pyannote speaker diarization running locally. The Whisper JAX endpoint is run asynchronously, meaning speaker diarization is run in parallel to the speech transcription. The diarized timestamps are aligned with the Whisper output to give the final speaker-segmented transcription.