Update app.py

app.py

@@ -5,6 +5,7 @@ import scipy.io.wavfile
 from transformers import VitsModel, AutoTokenizer
 import re
 
+# Load fine-tuned model from Hugging Face Hub or local path
 model = VitsModel.from_pretrained("Somali-tts/somali_tts_model")
 tokenizer = AutoTokenizer.from_pretrained("saleolow/somali-mms-tts")
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
@@ -75,16 +76,19 @@ def normalize_text(text):
     return text
 
 def tts(text):
-    ...
+    text = normalize_text(text)
+    inputs = tokenizer(text, return_tensors="pt").to(device)
+    with torch.no_grad():
+        waveform = model(**inputs).waveform.squeeze().cpu().numpy()
     filename = "output.wav"
     scipy.io.wavfile.write(filename, rate=model.config.sampling_rate, data=(waveform * 32767).astype(np.int16))
-    return filename  # ✅ This returns "output.wav"
+    return filename
 
 gr.Interface(
     fn=tts,
     inputs=gr.Textbox(label="Geli qoraal Soomaali ah"),
-    outputs=gr.Audio(type="filepath", label="Codka TTS"),  # ✅ filepath is required
+    outputs=gr.Audio(label="Codka TTS"),
     title="Somali TTS",
-    description="Ku qor qoraal Soomaali ah si aad u maqasho codka."
-).queue().launch()  # ✅ Enables API endpoint /run/predict
-
+    description="Ku qor qoraal Soomaaliyeed si aad u maqasho cod dabiici ah.",
+).launch()
+