Add custom handler for TTS inference

handler.py

@@ -66,9 +66,11 @@ class EndpointHandler:
         with torch.no_grad():
             out = self.model.generate(
                 **inputs,
-                speaker=speaker,  # Explicitly pass speaker to model
                 max_length=max_length,
                 output_audio=True,   # CSM returns audio
+                do_sample=True,
+                temperature=0.8,
+                top_p=0.9,
             )
 
         # Convert to base64 WAV for the endpoint response