Update app.py

app.py

@@ -5,14 +5,13 @@ import scipy.io.wavfile
 from transformers import VitsModel, AutoTokenizer
 import re
 
-# Load model
+# 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")
 model.to(device)
 model.eval()
 
-# Number normalization
 number_words = {
     0: "eber", 1: "koow", 2: "labo", 3: "seddex", 4: "afar", 5: "shan",
     6: "lix", 7: "todobo", 8: "sideed", 9: "sagaal", 10: "toban",
@@ -76,7 +75,6 @@ def normalize_text(text):
     text = text.replace("ZamZam", "SamSam")
     return text
 
-# TTS function
 def tts(text):
     text = normalize_text(text)
     inputs = tokenizer(text, return_tensors="pt").to(device)
@@ -84,13 +82,13 @@ def tts(text):
         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 f"https://HusseinBashir--Somali_tts.hf.space/file={filename}"
+    return filename
 
-# Interface + Launch
 gr.Interface(
     fn=tts,
     inputs=gr.Textbox(label="Geli qoraal Soomaali ah"),
-    outputs=gr.Audio(type="filepath", label="Codka TTS"),
-    title="Somali Text-to-Speech",
-    description="Ku qor qoraal Soomaali ah si aad u maqasho codka."
+    outputs=gr.Audio(label="Codka TTS"),
+    title="Somali TTS",
+    description="Ku qor qoraal Soomaaliyeed si aad u maqasho cod dabiici ah.",
 ).launch()
+