Update app.py

app.py

@@ -5,9 +5,11 @@ import scipy.io.wavfile
 from transformers import VitsModel, AutoTokenizer
 import re
 
-# Load fine-tuned model
-model = VitsModel.from_pretrained("./somali_tts_model")
+# Load fine-tuned model from Hugging Face Hub or local path
+model = VitsModel.from_pretrained("HusseinBashir/fine_tuned_vits_som")
 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_words = {
@@ -37,12 +39,10 @@ def number_to_words(number):
     elif number < 1000000:
         thousands, remainder = divmod(number, 1000)
         words = []
-
         if thousands == 1:
             words.append("kun")
         else:
             words.append(number_to_words(thousands) + " kun")
-
         if remainder >= 100:
             hundreds, rem2 = divmod(remainder, 100)
             if hundreds:
@@ -52,30 +52,24 @@ def number_to_words(number):
                 words.append("iyo " + number_to_words(rem2))
         elif remainder:
             words.append("iyo " + number_to_words(remainder))
-
         return " ".join(words)
     elif number < 1000000000:
         millions, remainder = divmod(number, 1000000)
         words = []
-
         if millions == 1:
             words.append("milyan")
         else:
             words.append(number_to_words(millions) + " milyan")
-
         if remainder:
             words.append(number_to_words(remainder))
-
         return " ".join(words)
     else:
         return str(number)
 
 def normalize_text(text):
-    # Convert numbers to Somali words
     numbers = re.findall(r'\d+', text)
     for num in numbers:
         text = text.replace(num, number_to_words(num))
-    # Replace foreign characters
     text = text.replace("KH", "qa").replace("Z", "S")
     text = text.replace("SH", "SHa'a").replace("DH", "Dha'a")
     text = text.replace("ZamZam", "SamSam")
@@ -83,7 +77,7 @@ def normalize_text(text):
 
 def tts(text):
     text = normalize_text(text)
-    inputs = tokenizer(text, return_tensors="pt")
+    inputs = tokenizer(text, return_tensors="pt").to(device)
     with torch.no_grad():
         waveform = model(**inputs).waveform.squeeze().cpu().numpy()
     filename = "output.wav"
@@ -96,4 +90,4 @@ gr.Interface(
     outputs=gr.Audio(label="Codka TTS"),
     title="Somali TTS",
     description="Ku qor qoraal Soomaaliyeed si aad u maqasho cod dabiici ah.",
-).launch()
+).launch()