Update app.py

app.py

@@ -1,4 +1,3 @@
-
 import gradio as gr
 import torch
 import torchaudio
@@ -12,8 +11,7 @@ import numpy as np
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
 # --- ADD ALL YOUR VOICE FILES HERE ---
-# The code will automatically create a dropdown for these files.
-# Make sure these files are in the same directory as your script.
+# Hubi in faylashan ay ku jiraan isla galka uu koodhkani ku jiro.
 VOICE_SAMPLE_FILES = ["1.wav", "1005.wav", "1060.wav", "737.wav"]
 
 # Directory to store speaker embedding files
@@ -21,7 +19,6 @@ EMBEDDING_DIR = "speaker_embeddings"
 os.makedirs(EMBEDDING_DIR, exist_ok=True)
 
 # --- Load Models ---
-# This part loads all the necessary AI models.
 try:
     print("Loading models... This may take a moment.")
     processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
@@ -36,15 +33,9 @@ try:
 except Exception as e:
     raise gr.Error(f"Error loading models: {e}. Check your internet connection.")
 
-# A dictionary to cache loaded speaker embeddings in memory
 speaker_embeddings_cache = {}
 
-# --- Function to Get or Create Speaker Embedding ---
 def get_speaker_embedding(wav_file_path):
-    """
-    Loads a speaker embedding from cache or file. If not found, creates and saves it.
-    """
-    # Check cache first
     if wav_file_path in speaker_embeddings_cache:
         return speaker_embeddings_cache[wav_file_path]
 
@@ -58,7 +49,7 @@ def get_speaker_embedding(wav_file_path):
 
     print(f"Creating new speaker embedding for {wav_file_path}...")
     if not os.path.exists(wav_file_path):
-        raise gr.Error(f"Audio file not found: {wav_file_path}. Please make sure it's in the correct directory.")
+        raise gr.Error(f"Audio file not found: {wav_file_path}.")
 
     try:
         audio, sr = torchaudio.load(wav_file_path)
@@ -76,10 +67,9 @@ def get_speaker_embedding(wav_file_path):
         print(f"Embedding created and saved for {wav_file_path}.")
         return embedding.to(device)
     except Exception as e:
-        raise gr.Error(f"Could not process audio file {wav_file_path}. Is it a valid WAV file? Error: {e}")
+        raise gr.Error(f"Could not process audio file {wav_file_path}. Error: {e}")
 
-# --- Text Processing Functions (Somali Number Conversion) ---
-# These functions remain the same.
+# --- Text Processing Functions (Remains the same) ---
 number_words = {
     0: "eber", 1: "kow", 2: "labo", 3: "saddex", 4: "afar", 5: "shan",
     6: "lix", 7: "toddobo", 8: "siddeed", 9: "sagaal", 10: "toban",
@@ -96,21 +86,16 @@ def number_to_words(n):
     if n < 1000: return (number_words[n//100] + " boqol" if n//100 > 1 else "boqol") + (" iyo " + number_to_words(n%100) if n%100 else "")
     if n < 1000000: return (number_to_words(n//1000) + " kun" if n//1000 > 1 else "kun") + (" iyo " + number_to_words(n%1000) if n%1000 else "")
     return str(n)
-
 def replace_numbers_with_words(text):
     return re.sub(r'\b\d+\b', lambda m: number_to_words(int(m.group())), text)
-
 def normalize_text(text):
     text = text.lower()
     text = replace_numbers_with_words(text)
     text = re.sub(r'[^\w\s\']', '', text)
     return text
 
-# --- Main Text-to-Speech Function ---
+# --- Main Text-to-Speech Function (WAXAAN KU DARNAY HAGAAJIN HAKAN) ---
 def text_to_speech(text, voice_choice):
-    """
-    Takes text and the chosen voice file, and returns audio.
-    """
     if not text:
         gr.Warning("Please enter some text.")
         return None, None
@@ -118,23 +103,26 @@ def text_to_speech(text, voice_choice):
         gr.Warning("Please select a voice from the dropdown.")
         return None, None
         
-    # Get the correct speaker embedding for the chosen voice
     speaker_embedding = get_speaker_embedding(voice_choice)
-
     normalized_text = normalize_text(text)
     inputs = processor(text=normalized_text, return_tensors="pt").to(device)
 
+    # Waa kan isbeddelka la sameeyay si codka loo hagaajiyo
     with torch.no_grad():
-        speech = model.generate_speech(
-            inputs["input_ids"],
+        speech = model.generate(
+            input_ids=inputs["input_ids"],
             speaker_embeddings=speaker_embedding.unsqueeze(0),
-            vocoder=vocoder
+            # Halbeegyada lagu daray si loo yareeyo dareenka AI-ga
+            do_sample=True, 
+            top_k=50,
         )
+        
+        # Vocoder-ka si gooni ah ayaa loo isticmaalayaa hadda
+        speech = vocoder(speech)
 
     return (16000, speech.cpu().numpy())
 
-# --- Gradio Interface ---
-# The user interface now includes a dropdown menu for voice selection.
+# --- Gradio Interface (Remains the same) ---
 iface = gr.Interface(
     fn=text_to_speech,
     inputs=[
@@ -143,7 +131,7 @@ iface = gr.Interface(
             VOICE_SAMPLE_FILES,
             label="Select Voice",
             info="Choose the voice you want to use for the speech.",
-            value=VOICE_SAMPLE_FILES[0] if VOICE_SAMPLE_FILES else None # Default to the first voice
+            value=VOICE_SAMPLE_FILES[0] if VOICE_SAMPLE_FILES else None
         )
     ],
     outputs=gr.Audio(label="Codka La Abuuray (Generated Voice)", type="numpy"),
@@ -155,12 +143,10 @@ iface = gr.Interface(
     ]
 )
 
-# Launch the web interface
 if __name__ == "__main__":
-    # Pre-load embeddings for a faster startup experience
     print("Pre-loading all voice embeddings...")
     for voice_file in VOICE_SAMPLE_FILES:
         get_speaker_embedding(voice_file)
     print("All voices are ready. Launching interface.")
     
-    iface.launch(share=True)
+    iface.launch(share=True)