Update app.py

app.py

@@ -1,11 +1,11 @@
 import torch  
 torch.manual_seed(160923)  
-  
+
 import gradio as gr  
 from huggingface_hub import hf_hub_download  
 from InferenceInterfaces.ControllableInterface import ControllableInterface  
 from Utility.utils import float2pcm, load_json_from_path  
-  
+
 import matplotlib.pyplot as plt  
 import librosa  
 import librosa.display  
@@ -13,7 +13,7 @@ import numpy as np
 import io  
 from PIL import Image  
 import threading  
-  
+
 def generate_spectrogram_image(wav, sr):  
     fig, ax = plt.subplots(figsize=(4, 1.5))  
     D = librosa.amplitude_to_db(librosa.stft(wav, n_fft=512), ref=np.max)  
@@ -24,9 +24,9 @@ def generate_spectrogram_image(wav, sr):
     plt.close(fig)  
     buf.seek(0)  
     return Image.open(buf)  
-  
+
 class TTSWebUI:  
-  
+
     def __init__(self,  
                  gpu_id="cpu",  
                  title="Controllable Text-to-Speech for over 7000 Languages",  
@@ -35,11 +35,11 @@ class TTSWebUI:
                  vocoder_model_path=None,  
                  embedding_gan_path=None,  
                  available_artificial_voices=10):  
-  
+
         path_to_iso_list = hf_hub_download(repo_id="Flux9665/ToucanTTS", filename="iso_to_fullname.json")  
         iso_to_name = load_json_from_path(path_to_iso_list)  
         self.text_selection = [f"{iso_to_name[iso]} ({iso})" for iso in iso_to_name]  
-  
+
         self.controllable_ui = ControllableInterface(  
             gpu_id=gpu_id,  
             available_artificial_voices=available_artificial_voices,  
@@ -47,11 +47,11 @@ class TTSWebUI:
             vocoder_model_path=vocoder_model_path,  
             embedding_gan_path=embedding_gan_path  
         )  
-  
+
         self.title = title  
         self.article = article  
         self.available_artificial_voices = available_artificial_voices  
-  
+
     def read(self,  
              prompt,  
              language,  
@@ -60,12 +60,13 @@ class TTSWebUI:
              voice_seed,  
              emb1,  
              reference_audio):  
-  
+
         lang_code = language[-4:-1]  
         result = [None]  
-  
+
         def run_tts():  
             try:  
+                print("[INFO] Running TTS with prompt:", prompt)  
                 result[0] = self.controllable_ui.read(  
                     prompt,  
                     reference_audio,  
@@ -80,23 +81,32 @@ class TTSWebUI:
                     -24.0  
                 )  
             except Exception as e:  
+                print("[ERROR] Exception during TTS:", e)  
                 result[0] = e  
-  
+
         thread = threading.Thread(target=run_tts)  
         thread.start()  
-        thread.join(timeout=20)  
-  
+        thread.join()  # ❗ Đã bỏ timeout để không bị cắt sớm trên CPU chậm
+
         if thread.is_alive():  
+            print("[WARNING] TTS thread still alive after join → Timeout logic (shouldn't happen now)")  
             return None, generate_spectrogram_image(np.zeros(16000), 16000)  
-  
+
         if isinstance(result[0], Exception):  
+            print("[ERROR] TTS returned exception object:", result[0])  
             raise result[0]  
-  
+
+        if result[0] is None:  
+            print("[ERROR] TTS returned None — possible silent failure")  
+            return None, generate_spectrogram_image(np.zeros(16000), 16000)  
+
         sr, wav, _ = result[0]  
-  
+
+        print("[INFO] TTS success — sample rate:", sr, " | waveform shape:", wav.shape)  
+
         fig = generate_spectrogram_image(wav, sr)  
         return (sr, float2pcm(wav)), fig  
-  
+
     def launch(self):  
         gr.Interface(  
             fn=self.read,  
@@ -118,7 +128,8 @@ class TTSWebUI:
             description=self.article,  
             theme=gr.themes.Ocean(primary_hue="amber", secondary_hue="orange")  
         ).launch()  
-  
+
 if __name__ == '__main__':  
     app = TTSWebUI(gpu_id="cpu")  
-    app.launch()
+    app.launch()
+