Create app.py

app.py

@@ -0,0 +1,63 @@
+import gradio as gr
+import torch
+import torchaudio
+
+# Load MMS-TTS components
+bundle = torchaudio.pipelines.MMS_TTS_ENG
+text_processor = bundle.get_text_processor()
+tacotron2 = bundle.get_tacotron2()
+waveglow = bundle.get_waveglow()
+
+# Set up device
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+tacotron2 = tacotron2.to(device)
+waveglow = waveglow.to(device)
+
+def synthesize_speech(text):
+    try:
+        if not text.strip():
+            raise ValueError("Text input cannot be empty")
+        
+        with torch.inference_mode():
+            # Process text input
+            processed, lengths = text_processor(text)
+            processed = processed.to(device)
+            lengths = lengths.to(device)
+            
+            # Generate spectrogram
+            spec, spec_lengths = tacotron2(processed, lengths)
+            
+            # Generate waveform
+            waveform, lengths = waveglow(spec, spec_lengths)
+            
+        # Convert to numpy array for Gradio
+        waveform = waveform.cpu().squeeze().numpy()
+        return (bundle.sample_rate, waveform)
+    
+    except Exception as e:
+        return f"Error: {str(e)}", None
+
+# Create Gradio interface
+interface = gr.Interface(
+    fn=synthesize_speech,
+    inputs=gr.Textbox(
+        label="Input Text",
+        placeholder="Enter text to synthesize...",
+        lines=3
+    ),
+    outputs=gr.Audio(
+        label="Generated Speech",
+        type="numpy"
+    ),
+    title="MMS-TTS English Text-to-Speech",
+    description="Convert text to speech using Facebook's MMS-TTS-ENG model",
+    examples=[
+        ["Hello! This is a text-to-speech demonstration."],
+        ["The quick brown fox jumps over the lazy dog."],
+        ["Natural language processing is fascinating!"]
+    ]
+)
+
+# Launch the application
+if __name__ == "__main__":
+    interface.launch(server_name="0.0.0.0" if torch.cuda.is_available() else None)