Update app.py

app.py

@@ -1,23 +1,21 @@
 import gradio as gr
 import torch
-from transformers import AutoProcessor, AutoModelForTextToSpeech
-import scipy.io.wavfile
-import tempfile
-
-model_id = "facebook/mms-tts-lis"
+import torchaudio
+from transformers import VitsModel, VitsTokenizer
 
 def tts_lisu(text):
+    model_id = "facebook/mms-tts-lis"
     try:
-        processor = AutoProcessor.from_pretrained(model_id)
-        model = AutoModelForTextToSpeech.from_pretrained(model_id)
-
-        inputs = processor(text=text, return_tensors="pt")
+        tokenizer = VitsTokenizer.from_pretrained(model_id)
+        model = VitsModel.from_pretrained(model_id)
+        inputs = tokenizer(text, return_tensors="pt")
         with torch.no_grad():
-            speech = model(**inputs).waveform
-
-        with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f:
-            scipy.io.wavfile.write(f.name, rate=16000, data=speech.squeeze().cpu().numpy())
-            return "✅ Giọng Lisu (ꓡꓲ-ꓢꓴ)", f.name
+            output = model(**inputs)
+        waveform = output.waveform.squeeze().cpu()
+        sample_rate = model.config.sampling_rate
+        path = "lisu.wav"
+        torchaudio.save(path, waveform.unsqueeze(0), sample_rate)
+        return "✅ Giọng Lisu (ꓡꓲ-ꓢꓴ)", path
     except Exception as e:
         return f"❌ Lỗi: {str(e)}", None