Update app.py

app.py

@@ -155,7 +155,7 @@ def predict(text, speaker):
     input_ids = inputs["input_ids"]
     input_ids = input_ids[..., :model.config.max_text_positions]
 
-    speaker_embedding = np.load(speaker_embeddings[speaker[:3]])
+    speaker_embedding = np.load(speaker_embeddings[speaker[:3]]).astype(np.float32)
 
     speaker_embedding = torch.tensor(speaker_embedding).unsqueeze(0)