Update app.py

app.py

@@ -6,25 +6,23 @@ import random
 import torch
 import openai
 
-# Check if running in a duplicate space
 IS_DUPLICATE = not os.getenv('SPACE_ID', '').startswith('hexgrad/')
 CHAR_LIMIT = None if IS_DUPLICATE else 5000
 
-# Check if CUDA is available
 CUDA_AVAILABLE = torch.cuda.is_available()
-
-# Load the models (GPU and CPU versions)
 models = {gpu: KModel().to('cuda' if gpu else 'cpu').eval() for gpu in [False] + ([True] if CUDA_AVAILABLE else [])}
 pipelines = {lang_code: KPipeline(lang_code=lang_code, model=False) for lang_code in 'abefhijpz'}
 pipelines['a'].g2p.lexicon.golds['kokoro'] = 'kˈOkəɹO'
 pipelines['b'].g2p.lexicon.golds['kokoro'] = 'kˈQkəɹQ'
 
-# GPU function to generate audio
+# Check API status
+API_OPEN = os.getenv('SPACE_ID') != 'hexgrad/Kokoro-TTS'
+API_NAME = None if API_OPEN else False
+
 @spaces.GPU(duration=10)
 def forward_gpu(ps, ref_s, speed):
     return models[True](ps, ref_s, speed)
 
-# Function to generate first output
 def generate_first(text, voice='af_heart', speed=1, use_gpu=CUDA_AVAILABLE):
     text = text if CHAR_LIMIT is None else text.strip()[:CHAR_LIMIT]
     pipeline = pipelines[voice[0]]
@@ -47,17 +45,10 @@ def generate_first(text, voice='af_heart', speed=1, use_gpu=CUDA_AVAILABLE):
         return (24000, audio.numpy()), ps
     return None, ''
 
-# Function to tokenize first
 def tokenize_first(text, voice='af_heart'):
     words = text.split()  # This splits the text into words based on spaces
     return words  # Return a list of words
 
-# Function to get random text for the "Random Text" button
-random_texts = {}
-for lang in ['en']:
-    with open(f'{lang}.txt', 'r') as r:
-        random_texts[lang] = [line.strip() for line in r]
-
 def get_random_text(voice):
     lang = dict(a='en', b='en')[voice[0]]
     return random.choice(random_texts[lang])
@@ -81,7 +72,12 @@ def translate_and_generate(text, voice, speed):
     audio, tokens = generate_first(translated_text, voice, speed, use_gpu=CUDA_AVAILABLE)
     return audio, tokens, translated_text
 
-# Predefined voices for the dropdown menu
+# Load random text for "Random Text" button
+random_texts = {}
+for lang in ['en']:
+    with open(f'{lang}.txt', 'r') as r:
+        random_texts[lang] = [line.strip() for line in r]
+
 CHOICES = {
     '🇺🇸 🚺 Heart ❤️': 'af_heart',
     '🇺🇸 🚺 Bella 🔥': 'af_bella',
@@ -94,7 +90,7 @@ CHOICES = {
     '🇺🇸 🚺 Alloy': 'af_alloy',
     '🇺🇸 🚺 Jessica': 'af_jessica',
     '🇺🇸 🚺 River': 'af_river',
-    
+        
     '🇺🇸 🚹 Michael': 'am_michael',
     '🇺🇸 🚹 Fenrir': 'am_fenrir',
     '🇺🇸 🚹 Puck': 'am_puck',
@@ -104,62 +100,62 @@ CHOICES = {
     '🇺🇸 🚹 Onyx': 'am_onyx',
     '🇺🇸 🚹 Santa': 'am_santa',
     '🇺🇸 🚹 Adam': 'am_adam',
-    
+        
     '🇬🇧 🚺 Emma': 'bf_emma',
     '🇬🇧 🚺 Isabella': 'bf_isabella',
     '🇬🇧 🚺 Alice': 'bf_alice',
     '🇬🇧 🚺 Lily': 'bf_lily',
-    
+        
     '🇬🇧 🚹 George': 'bm_george',
     '🇬🇧 🚹 Fable': 'bm_fable',
     '🇬🇧 🚹 Lewis': 'bm_lewis',
     '🇬🇧 🚹 Daniel': 'bm_daniel',
-    
+        
     '🇪🇸 🚺 Dora': 'ef_dora',
-    
+        
     '🇪🇸 🚹 Alex': 'em_alex',
     '🇪🇸 🚹 Santa': 'em_santa',
-    
+        
     '🇫🇷 🚺 Siwis': 'ff_siwis',
-    
+        
     '🇮🇳 🚹 Alpha': 'hf_alpha',
     '🇮🇳 🚹 Beta': 'hf_beta',
-    
+        
     '🇮🇳 🚹 Omega': 'hm_omega',
     '🇮🇳 🚹 Psi': 'hm_psi',
-    
+        
     '🇮🇹 🚺 Sara': 'if_sara',
-    
+        
     '🇮🇹 🚺 Nicola': 'im_nicola',
-    
+        
     '🇯🇵 🚹 Alpha': 'jf_alpha',
     '🇯🇵 🚹 Gongitsune': 'jf_gongitsune',
     '🇯🇵 🚹 Nezumi': 'jf_nezumi',
     '🇯🇵 🚹 Tebukuro': 'jf_tebukuro',
-    
+        
     '🇯🇵 🚹 Kumo': 'jm_kumo',
-    
+        
     '🇧🇷 🚺 Dora': 'pf_dora',
-    
+        
     '🇧🇷 🚹 Alex': 'pm_alex',
     '🇧🇷 🚹 Santa': 'pm_santa',
-    
+        
     '🇨🇳 🚺 Xiaobei': 'zf_xiaobei',
     '🇨🇳 🚺 Xiaoni': 'zf_xiaoni',
     '🇨🇳 🚺 Xiaoxiao': 'zf_xiaoxiao',
     '🇨🇳 🚺 Xiaoyi': 'zf_xiaoyi',
-
+    
     '🇨🇳 🚹 Yunjian': 'zm_yunjian',
     '🇨🇳 🚹 Yunxi': 'zm_yunxi',
     '🇨🇳 🚹 Yunxia': 'zm_yunxia',
     '🇨🇳 🚹 Yunyang': 'zm_yunyang',
+    # (All the voice options here... same as before)
 }
 
-# Load voices
 for v in CHOICES.values():
     pipelines[v[0]].load_voice(v)
 
-# Build the interface
+# Tabs for generation and translation
 with gr.Blocks() as generate_tab:
     out_audio = gr.Audio(label='Output Audio', interactive=False, streaming=False, autoplay=True)
     generate_btn = gr.Button('Generate', variant='primary')
@@ -168,21 +164,22 @@ with gr.Blocks() as generate_tab:
         tokenize_btn = gr.Button('Tokenize', variant='secondary')
         predict_btn = gr.Button('Predict', variant='secondary', visible=False)
 
-# Translator Tab
 with gr.Blocks() as translator_tab:
     trans_out_audio = gr.Audio(label='Translated Audio Output', interactive=False, streaming=False, autoplay=True)
     trans_out_tokens = gr.Textbox(interactive=False, show_label=False, info='Tokens used to generate the translated audio')
     translate_btn = gr.Button('Translate & Generate Audio', variant='primary')
     
-    translate_btn.click(fn=translate_and_generate, inputs=[text, voice, speed], outputs=[trans_out_audio, trans_out_tokens, text], api_name=None)
+    translate_btn.click(fn=translate_and_generate, inputs=[text, voice, speed], outputs=[trans_out_audio, trans_out_tokens, text], api_name=API_NAME)
+
+BANNER_TEXT = '''
+[***Kokoro*** **is an open-weight TTS model with 82 million parameters.**](https://huggingface.co/hexgrad/Kokoro-82M)
+As of January 31st, 2025, Kokoro was the most-liked [**TTS model**](https://huggingface.co/models?pipeline_tag=text-to-speech&sort=likes) and the most-liked [**TTS space**](https://huggingface.co/spaces?sort=likes&search=tts) on Hugging Face.
+This demo only showcases English, but you can directly use the model to access other languages.
+'''
 
-# Main Interface
 with gr.Blocks() as app:
     with gr.Row():
-        gr.Markdown('''[***Kokoro*** **is an open-weight TTS model with 82 million parameters.**](https://huggingface.co/hexgrad/Kokoro-82M)
-        As of January 31st, 2025, Kokoro was the most-liked [**TTS model**](https://huggingface.co/models?pipeline_tag=text-to-speech&sort=likes) and the most-liked [**TTS space**](https://huggingface.co/spaces?sort=likes&search=tts) on Hugging Face.
-        This demo only showcases English, but you can directly use the model to access other languages.''', container=True)
-    
+        gr.Markdown(BANNER_TEXT, container=True)
     with gr.Row():
         with gr.Column():
             text = gr.Textbox(label='Input Text', info=f"Up to ~500 characters per Generate, or {'∞' if CHAR_LIMIT is None else CHAR_LIMIT} characters per Stream")
@@ -200,10 +197,10 @@ with gr.Blocks() as app:
         with gr.Column():
             gr.TabbedInterface([generate_tab, translator_tab], ['Generate', 'Translator'])
     
-    random_btn.click(fn=get_random_text, inputs=[voice], outputs=[text])
-    generate_btn.click(fn=generate_first, inputs=[text, voice, speed, use_gpu], outputs=[out_audio, out_ps])
-    tokenize_btn.click(fn=tokenize_first, inputs=[text, voice], outputs=[out_ps])
-    predict_btn.click(fn=predict, inputs=[text, voice, speed], outputs=[out_audio])
+    random_btn.click(fn=get_random_text, inputs=[voice], outputs=[text], api_name=API_NAME)
+    generate_btn.click(fn=generate_first, inputs=[text, voice, speed, use_gpu], outputs=[out_audio, out_ps], api_name=API_NAME)
+    tokenize_btn.click(fn=tokenize_first, inputs=[text, voice], outputs=[out_ps], api_name=API_NAME)
+    predict_btn.click(fn=predict, inputs=[text, voice, speed], outputs=[out_audio], api_name=API_NAME)
 
 if __name__ == '__main__':
-    app.queue().launch(show_api=True)
+    app.queue(api_open=API_OPEN).launch(show_api=API_OPEN, ssr_mode=True)