Update app.py

app.py

@@ -1,3 +1,47 @@
+import torch
+from transformers import AutoModelForTextToSpeech, AutoProcessor
+import soundfile as sf  # For saving the audio
 import gradio as gr
 
-gr.Interface.load("models/facebook/fastspeech2-en-ljspeech").launch()
+# 1. Choose the model and processor
+model_name = "facebook/fastspeech2-en-ljspeech"
+
+# 2. Load the processor and model
+processor = AutoProcessor.from_pretrained(model_name)
+model = AutoModelForTextToSpeech.from_pretrained(model_name)
+
+# 3. Move the model to the GPU (if available)
+device = "cuda" if torch.cuda.is_available() else "cpu"
+model = model.to(device)
+
+# 4. Define a function for text-to-speech
+def synthesize_speech(text):
+    try:
+        inputs = processor(text=text, return_tensors="pt")
+        # Move input tensors to the same device as the model
+        inputs = {key: value.to(device) for key, value in inputs.items()}
+        with torch.no_grad():  # Disable gradient calculation during inference
+            output = model(**inputs).waveform
+        # Move to cpu before converting
+        output = output.cpu()
+
+        # Convert the output to a NumPy array (required by soundfile)
+        waveform = output.squeeze().numpy()
+
+        # Return the waveform and the sample rate (needed for Gradio)
+        return (processor.feature_extractor.sampling_rate, waveform)
+    except Exception as e:
+      print (e)
+      return (None, None) # in case of error
+
+# 5. create interface
+iface = gr.Interface(
+    fn=synthesize_speech,
+    inputs=gr.Textbox(lines=2, placeholder="Enter text here..."),
+    outputs=gr.Audio(label="Generated Speech", type="numpy"),
+    title="FastSpeech2 Text-to-Speech",
+    description="Enter text to synthesize speech using FastSpeech2.",
+)
+
+# 6. launch
+iface.launch()