Create app.py

app.py

@@ -0,0 +1,97 @@
+import gradio as gr
+import torch
+import numpy as np
+from transformers import pipeline
+
+# 1. Setup device (Use GPU if available on the Space, otherwise CPU)
+device = "cuda:0" if torch.cuda.is_available() else "cpu"
+
+print(f"Loading aoxo/swaram model on {device}...")
+
+# 2. Load the TTS pipeline globally so it only loads once when the Space starts
+try:
+    synthesizer = pipeline("text-to-speech", model="aoxo/swaram", device=device)
+    print("Model loaded successfully!")
+except Exception as e:
+    print(f"Error loading model: {e}")
+    synthesizer = None
+
+# 3. Define the prediction function
+def generate_audio(text):
+    if not text.strip():
+        return None, "Please enter some text."
+        
+    if synthesizer is None:
+        return None, "Error: Model failed to load. Check Space logs."
+
+    try:
+        # Generate speech
+        speech = synthesizer(text)
+        
+        # The transformers pipeline returns a dictionary: 
+        # {'audio': numpy array, 'sampling_rate': int}
+        audio_data = speech["audio"]
+        sample_rate = speech["sampling_rate"]
+        
+        # Gradio expects audio in (sample_rate, numpy_1D_array) format
+        # Pipeline audio is usually shape (1, N). We squeeze it to (N,)
+        if len(audio_data.shape) > 1:
+            audio_data = np.squeeze(audio_data)
+            
+        return (sample_rate, audio_data), "Success!"
+        
+    except Exception as e:
+        return None, f"Generation Error: {str(e)}"
+
+# 4. Build the Gradio Interface
+with gr.Blocks(title="Swaram Malayalam TTS", theme=gr.themes.Soft()) as demo:
+    gr.Markdown(
+        """
+        # 🗣️ Swaram Malayalam Text-to-Speech
+        Enter Malayalam text below to generate speech using the `aoxo/swaram` model.
+        """
+    )
+    
+    with gr.Row():
+        with gr.Column():
+            text_input = gr.Textbox(
+                label="Enter Malayalam Text",
+                placeholder="മലയാളം ടൈപ്പ് ചെയ്യുക...",
+                lines=5
+            )
+            
+            with gr.Row():
+                clear_btn = gr.Button("Clear")
+                generate_btn = gr.Button("Generate Speech", variant="primary")
+            
+            gr.Examples(
+                examples=[
+                    ["നമസ്കാരം, ഇതെന്റെ പുതിയ ശബ്ദമാണ്."],
+                    ["കേരളം ദൈവത്തിന്റെ സ്വന്തം നാടാണ്."],
+                    ["കള്ളാ കടയാടി മോനെ"]
+                ],
+                inputs=[text_input],
+                label="Examples"
+            )
+            
+        with gr.Column():
+            audio_output = gr.Audio(label="Generated Audio", type="numpy", interactive=False)
+            status_output = gr.Textbox(label="Status", interactive=False)
+            
+    # Event Listeners
+    generate_btn.click(
+        fn=generate_audio,
+        inputs=[text_input],
+        outputs=[audio_output, status_output],
+        api_name="synthesize" # Allows this Space to be used as an API later
+    )
+    
+    clear_btn.click(
+        fn=lambda: (None, None, ""),
+        inputs=[],
+        outputs=[text_input, audio_output, status_output]
+    )
+
+# 5. Launch the app
+if __name__ == "__main__":
+    demo.launch()