Update app.py

app.py

@@ -1,38 +1,44 @@
 import gradio as gr
 import torch
-from transformers import AutoProcessor, BarkModel
-from scipy.io.wavfile import write
+from transformers import VitsModel, AutoTokenizer
+import numpy as np
+import scipy.io.wavfile
 import tempfile
 
-# Load model and processor (Free Hugging Face model)
-processor = AutoProcessor.from_pretrained("suno/bark-small")
-model = BarkModel.from_pretrained("suno/bark-small")
+# Load lightweight MMS TTS model
+model = VitsModel.from_pretrained("facebook/mms-tts-eng")
+tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-eng")
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
 model = model.to(device)
 
 def text_to_speech(text):
-    inputs = processor(text, return_tensors="pt").to(device)
-    
+    if text.strip() == "":
+        return None
+
+    inputs = tokenizer(text, return_tensors="pt").to(device)
+
     with torch.no_grad():
-        audio_array = model.generate(**inputs)
-    
-    audio_array = audio_array.cpu().numpy().squeeze()
+        output = model(**inputs).waveform
+
+    audio = output.squeeze().cpu().numpy()
     
-    # Save temporary WAV file
-    sample_rate = model.generation_config.sample_rate
+    # Normalize audio
+    audio = audio / np.max(np.abs(audio))
+
+    # Save to temporary WAV file
+    sample_rate = model.config.sampling_rate
     temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=".wav")
-    write(temp_file.name, sample_rate, audio_array)
-    
+    scipy.io.wavfile.write(temp_file.name, sample_rate, audio.astype(np.float32))
+
     return temp_file.name
 
-# Gradio Interface
 demo = gr.Interface(
     fn=text_to_speech,
     inputs=gr.Textbox(label="Enter Text"),
     outputs=gr.Audio(label="Generated Speech"),
-    title="Simple Text-to-Speech App",
-    description="Enter text and generate speech using a free Hugging Face model."
+    title="Fast Text-to-Speech (Lightweight)",
+    description="Fast and accurate TTS using facebook/mms-tts-eng"
 )
 
 if __name__ == "__main__":