Update app.py

app.py

@@ -1,33 +1,12 @@
+from transformers import AutoProcessor, VitsModel
 import torch
-from transformers import VITSModel, AutoProcessor
-import gradio as gr
-import scipy.io.wavfile
+import soundfile as sf
 
-# Load model and processor
-model = VITSModel.from_pretrained("facebook/mms-tts-hmn")
 processor = AutoProcessor.from_pretrained("facebook/mms-tts-hmn")
+model = VitsModel.from_pretrained("facebook/mms-tts-hmn")
 
-# Set model to eval mode
-model.eval()
+inputs = processor(text="Kuv hlub koj", return_tensors="pt")
+with torch.no_grad():
+    speech = model(**inputs).waveform
 
-def tts(text):
-    # Preprocess input
-    inputs = processor(text, return_tensors="pt")
-
-    # Generate audio
-    with torch.no_grad():
-        output = model(**inputs)
-    audio = output.waveform[0].numpy()
-
-    # Convert to 16-bit PCM WAV for Gradio playback
-    sample_rate = model.config.sampling_rate
-    return (sample_rate, audio)
-
-# Gradio UI
-gr.Interface(
-    fn=tts,
-    inputs=gr.Textbox(label="Nhập văn bản tiếng H'Mông"),
-    outputs=gr.Audio(label="Phát âm"),
-    title="Text-to-Speech tiếng H'Mông (TTS)",
-    description="TTS sử dụng mô hình facebook/mms-tts-hmn (Meta MMS VITS)."
-).launch()
+sf.write("output.wav", speech.numpy()[0], samplerate=16000)