Update app.py

app.py

@@ -4,7 +4,6 @@ import gradio as gr
 import os
 import random
 import torch
-import openai
 
 IS_DUPLICATE = not os.getenv('SPACE_ID', '').startswith('hexgrad/')
 CHAR_LIMIT = None if IS_DUPLICATE else 5000
@@ -15,10 +14,6 @@ pipelines = {lang_code: KPipeline(lang_code=lang_code, model=False) for lang_cod
 pipelines['a'].g2p.lexicon.golds['kokoro'] = 'kˈOkəɹO'
 pipelines['b'].g2p.lexicon.golds['kokoro'] = 'kˈQkəɹQ'
 
-# 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)
@@ -50,116 +45,115 @@ def predict(text, voice='af_heart', speed=1):
     return generate_first(text, voice, speed, use_gpu=False)[0]
 
 def tokenize_first(text, voice='af_heart'):
+    # Split the input text into words and return as a list of words (fix applied here)
     words = text.split()  # This splits the text into words based on spaces
     return words  # Return a list of words
 
-def get_random_text(voice):
-    lang = dict(a='en', b='en')[voice[0]]
-    return random.choice(random_texts[lang])
-
-# OpenAI GPT-4 translation function
-def translate_to_english(text, model="gpt-4"):
-    try:
-        response = openai.Completion.create(
-            model=model,
-            prompt=f"Translate the following text to English:\n\n{text}",
-            temperature=0.5,
-            max_tokens=500,
-        )
-        return response.choices[0].text.strip()
-    except Exception as e:
-        return str(e)
-
-# Function to handle generation for translated text
-def translate_and_generate(text, voice, speed):
-    translated_text = translate_to_english(text)
-    audio, tokens = generate_first(translated_text, voice, speed, use_gpu=CUDA_AVAILABLE)
-    return audio, tokens, translated_text
+def generate_all(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]]
+    pack = pipeline.load_voice(voice)
+    use_gpu = use_gpu and CUDA_AVAILABLE
+    for _, ps, _ in pipeline(text, voice, speed):
+        ref_s = pack[len(ps)-1]
+        try:
+            if use_gpu:
+                audio = forward_gpu(ps, ref_s, speed)
+            else:
+                audio = models[False](ps, ref_s, speed)
+        except gr.exceptions.Error as e:
+            if use_gpu:
+                gr.Warning(str(e))
+                gr.Info('Switching to CPU')
+                audio = models[False](ps, ref_s, speed)
+            else:
+                raise gr.Error(e)
+        yield 24000, audio.numpy()
 
-# 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]
 
+def get_random_text(voice):
+    lang = dict(a='en', b='en')[voice[0]]
+    return random.choice(random_texts[lang])
+
 CHOICES = {
-    '🇺🇸 🚺 Heart ❤️': 'af_heart',
-    '🇺🇸 🚺 Bella 🔥': 'af_bella',
-    '🇺🇸 🚺 Nicole 🎧': 'af_nicole',
-    '🇺🇸 🚺 Aoede': 'af_aoede',
-    '🇺🇸 🚺 Kore': 'af_kore',
-    '🇺🇸 🚺 Sarah': 'af_sarah',
-    '🇺🇸 🚺 Nova': 'af_nova',
-    '🇺🇸 🚺 Sky': 'af_sky',
-    '🇺🇸 🚺 Alloy': 'af_alloy',
-    '🇺🇸 🚺 Jessica': 'af_jessica',
-    '🇺🇸 🚺 River': 'af_river',
-        
-    '🇺🇸 🚹 Michael': 'am_michael',
-    '🇺🇸 🚹 Fenrir': 'am_fenrir',
-    '🇺🇸 🚹 Puck': 'am_puck',
-    '🇺🇸 🚹 Echo': 'am_echo',
-    '🇺🇸 🚹 Eric': 'am_eric',
-    '🇺🇸 🚹 Liam': 'am_liam',
-    '🇺🇸 🚹 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',
+'🇺🇸 🚺 Heart ❤️': 'af_heart',
+'🇺🇸 🚺 Bella 🔥': 'af_bella',
+'🇺🇸 🚺 Nicole 🎧': 'af_nicole',
+'🇺🇸 🚺 Aoede': 'af_aoede',
+'🇺🇸 🚺 Kore': 'af_kore',
+'🇺🇸 🚺 Sarah': 'af_sarah',
+'🇺🇸 🚺 Nova': 'af_nova',
+'🇺🇸 🚺 Sky': 'af_sky',
+'🇺🇸 🚺 Alloy': 'af_alloy',
+'🇺🇸 🚺 Jessica': 'af_jessica',
+'🇺🇸 🚺 River': 'af_river',
+    
+'🇺🇸 🚹 Michael': 'am_michael',
+'🇺🇸 🚹 Fenrir': 'am_fenrir',
+'🇺🇸 🚹 Puck': 'am_puck',
+'🇺🇸 🚹 Echo': 'am_echo',
+'🇺🇸 🚹 Eric': 'am_eric',
+'🇺🇸 🚹 Liam': 'am_liam',
+'🇺🇸 🚹 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',
     
-    '🇨🇳 🚹 Yunjian': 'zm_yunjian',
-    '🇨🇳 🚹 Yunxi': 'zm_yunxi',
-    '🇨🇳 🚹 Yunxia': 'zm_yunxia',
-    '🇨🇳 🚹 Yunyang': 'zm_yunyang',
-    # (All the voice options here... same as before)
+'🇮🇹 🚺 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',
 }
-
 for v in CHOICES.values():
     pipelines[v[0]].load_voice(v)
 
-# 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')
@@ -168,24 +162,19 @@ with gr.Blocks() as generate_tab:
         tokenize_btn = gr.Button('Tokenize', variant='secondary')
         predict_btn = gr.Button('Predict', variant='secondary', visible=False)
 
-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')
-
 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 app with text input field that is accessible globally
+API_OPEN = os.getenv('SPACE_ID') != 'hexgrad/Kokoro-TTS'
+API_NAME = None if API_OPEN else False
 with gr.Blocks() as app:
     with gr.Row():
         gr.Markdown(BANNER_TEXT, container=True)
     with gr.Row():
         with gr.Column():
-            # Make 'text' globally accessible
             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")
             with gr.Row():
                 voice = gr.Dropdown(list(CHOICES.items()), value='af_heart', label='Voice', info='Quality and availability vary by language')
@@ -199,17 +188,11 @@ with gr.Blocks() as app:
             speed = gr.Slider(minimum=0.5, maximum=2, value=1, step=0.1, label='Speed')
             random_btn = gr.Button('Random Text', variant='secondary')
         with gr.Column():
-            # Tabs for generation and translation
-            gr.TabbedInterface([generate_tab, translator_tab], ['Generate', 'Translator'])
-    
-    # Event handlers for the buttons
+            gr.TabbedInterface([generate_tab], ['Generate'])
     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)
 
-    # Fix: Pass `text` as an input to `translate_and_generate`
-    translate_btn.click(fn=translate_and_generate, inputs=[text, voice, speed], outputs=[trans_out_audio, trans_out_tokens, text], api_name=API_NAME)
-
 if __name__ == '__main__':
-    app.queue(api_open=API_OPEN).launch(show_api=API_OPEN, ssr_mode=True)
+    app.queue(api_open=API_OPEN).launch(show_api=API_OPEN, ssr_mode=True)