Update app.py

app.py

@@ -11,8 +11,6 @@ model = SpeechT5ForTextToSpeech.from_pretrained(checkpoint)
 vocoder = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
 
 def predict(text, speaker):
-    if len(text.strip()) == 0:
-        return (16000, np.zeros(0).astype(np.int16))
 
     inputs = processor(text=text, return_tensors="pt")
 
@@ -20,24 +18,6 @@ def predict(text, speaker):
     input_ids = inputs["input_ids"]
     input_ids = input_ids[..., :model.config.max_text_positions]
 
-    if speaker == "Surprise Me!":
-        # load one of the provided speaker embeddings at random
-        idx = np.random.randint(len(speaker_embeddings))
-        key = list(speaker_embeddings.keys())[idx]
-        speaker_embedding = np.load(speaker_embeddings[key])
-
-        # randomly shuffle the elements
-        np.random.shuffle(speaker_embedding)
-
-        # randomly flip half the values
-        x = (np.random.rand(512) >= 0.5) * 1.0
-        x[x == 0] = -1.0
-        speaker_embedding *= x
-
-        #speaker_embedding = np.random.rand(512).astype(np.float32) * 0.3 - 0.15
-    else:
-        speaker_embedding = np.load(speaker_embeddings[speaker[:3]])
-
     speaker_embedding = torch.tensor(speaker_embedding).unsqueeze(0)
 
     speech = model.generate_speech(input_ids, speaker_embedding, vocoder=vocoder)