Update app.py

app.py

@@ -2,32 +2,29 @@ import torch
 from transformers import VitsModel, AutoTokenizer
 import gradio as gr
 
-# Load the fine-tuned model and tokenizer
+# Load fine-tuned model and tokenizer
 model = VitsModel.from_pretrained("HusseinBashir/codad_tijaabo")
 tokenizer = AutoTokenizer.from_pretrained("HusseinBashir/codad_tijaabo")
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-model.to(device)
+model.to(device).eval()
 
-# Gradio TTS function
-# Gradio TTS function
 def tts(text):
     inputs = tokenizer(text, return_tensors="pt").to(device)
     with torch.no_grad():
-        output = model(**inputs).waveform.squeeze(1).cpu().numpy()
+        output = model(**inputs)
+        waveform = output["waveform"].squeeze(1).cpu().numpy()
 
-    # Ensure the output is a 1D numpy array and normalized
-    if output.ndim > 1:
-        output = output.flatten()
-    output = output / max(abs(output))  # Normalize to [-1, 1]
-    return (22050, output)  # Return a tuple (sample_rate, waveform)
+    # Normalize waveform
+    waveform = waveform.flatten()
+    waveform = waveform / max(abs(waveform))
 
+    return (22050, waveform)  # 22.05 kHz sample rate typical for VITS
 
-# Gradio interface for the TTS model
-iface = gr.Interface(
+# Gradio interface
+gr.Interface(
     fn=tts,
-    inputs=gr.Textbox(label="Enter text"),
-    outputs=gr.Audio(label="Generated Speech"),
-    title="Fine-tuned VITS TTS",
-    description="Generate speech from text using the fine-tuned VITS model."
-)
-iface.launch()
+    inputs=gr.Textbox(label="Geli qoraal Soomaali ah"),
+    outputs=gr.Audio(label="Codka la sameeyey"),
+    title="Codad Tijaabo TTS",
+    description="Ku qor qoraal Soomaali ah si aad cod u maqasho iyadoo la adeegsanayo VITS."
+).launch()