Update app.py

app.py

@@ -29,23 +29,12 @@ voicelist = ['f-us-1', 'f-us-2', 'f-us-3', 'f-us-4', 'm-us-1', 'm-us-2', 'm-us-3
 voices = {}
 import phonemizer
 global_phonemizer = phonemizer.backend.EspeakBackend(language='en-us', preserve_punctuation=True,  with_stress=True)
-# todo: cache computed style, load using pickle
-# if os.path.exists('voices.pkl'):
-    # with open('voices.pkl', 'rb') as f:
-        # voices = pickle.load(f)
-# else:
+
 for v in voicelist:
     voices[v] = styletts2importable.compute_style(f'voices/{v}.wav')
-# def synthesize(text, voice, multispeakersteps):
-#     if text.strip() == "":
-#         raise gr.Error("You must enter some text")
-#     # if len(global_phonemizer.phonemize([text])) > 300:
-#     if len(text) > 300:
-#         raise gr.Error("Text must be under 300 characters")
-#     v = voice.lower()
-#     # return (24000, styletts2importable.inference(text, voices[v], alpha=0.3, beta=0.7, diffusion_steps=7, embedding_scale=1))
-#     return (24000, styletts2importable.inference(text, voices[v], alpha=0.3, beta=0.7, diffusion_steps=multispeakersteps, embedding_scale=1))
+
 if not torch.cuda.is_available(): INTROTXT += "\n\n### You are on a CPU-only system, inference will be much slower.\n\nYou can use the [online demo](https://huggingface.co/spaces/styletts2/styletts2) for fast inference."
+
 def synthesize(text, voice, lngsteps, password, progress=gr.Progress()):
     if text.strip() == "":
         raise gr.Error("You must enter some text")
@@ -61,30 +50,8 @@ def synthesize(text, voice, lngsteps, password, progress=gr.Progress()):
         print(t)
         audios.append(styletts2importable.inference(t, voices[v], alpha=0.3, beta=0.7, diffusion_steps=lngsteps, embedding_scale=1))
     return (24000, np.concatenate(audios))
-# def longsynthesize(text, voice, lngsteps, password, progress=gr.Progress()):
-#     if password == os.environ['ACCESS_CODE']:
-#         if text.strip() == "":
-#             raise gr.Error("You must enter some text")
-#         if lngsteps > 25:
-#             raise gr.Error("Max 25 steps")
-#         if lngsteps < 5:
-#             raise gr.Error("Min 5 steps")
-#         texts = split_and_recombine_text(text)
-#         v = voice.lower()
-#         audios = []
-#         for t in progress.tqdm(texts):
-#             audios.append(styletts2importable.inference(t, voices[v], alpha=0.3, beta=0.7, diffusion_steps=lngsteps, embedding_scale=1))
-#         return (24000, np.concatenate(audios))
-#     else:
-#         raise gr.Error('Wrong access code')
+
 def rn_clsynthesize(text, voice, vcsteps, embscale, alpha, beta, progress=gr.Progress()):
-    # if text.strip() == "":
-    #     raise gr.Error("You must enter some text")
-    # # if global_phonemizer.phonemize([text]) > 300:
-    # if len(text) > 400:
-    #     raise gr.Error("Text must be under 400 characters")
-    # # return (24000, styletts2importable.inference(text, styletts2importable.compute_style(voice), alpha=0.3, beta=0.7, diffusion_steps=20, embedding_scale=1))
-    # return (24000, styletts2importable.inference(text, styletts2importable.compute_style(voice), alpha=0.3, beta=0.7, diffusion_steps=vcsteps, embedding_scale=1))
     if text.strip() == "":
         raise gr.Error("You must enter some text")
     if len(text) > 50000:
@@ -96,21 +63,13 @@ def rn_clsynthesize(text, voice, vcsteps, embscale, alpha, beta, progress=gr.Pro
     print("*** end ***")
     texts = txtsplit(text)
     audios = []
-    # vs = styletts2importable.compute_style(voice)
     vs = styletts2importable.compute_style(voice)
-    # print(vs)
     for t in progress.tqdm(texts):
         audios.append(styletts2importable.inference(t, vs, alpha=alpha, beta=beta, diffusion_steps=vcsteps, embedding_scale=embscale))
-        # audios.append(styletts2importable.inference(t, vs, diffusion_steps=10, alpha=0.3, beta=0.7, embedding_scale=5))
     return (24000, np.concatenate(audios))
+
 def rn_ljsynthesize(text, steps, progress=gr.Progress()):
-    # if text.strip() == "":
-    #     raise gr.Error("You must enter some text")
-    # # if global_phonemizer.phonemize([text]) > 300:
-    # if len(text) > 400:
-    #     raise gr.Error("Text must be under 400 characters")
     noise = torch.randn(1,1,256).to('cuda' if torch.cuda.is_available() else 'cpu')
-    # return (24000, ljspeechimportable.inference(text, noise, diffusion_steps=7, embedding_scale=1))
     if text.strip() == "":
         raise gr.Error("You must enter some text")
     if len(text) > 150000:
@@ -124,58 +83,58 @@ def rn_ljsynthesize(text, steps, progress=gr.Progress()):
         audios.append(ljspeechimportable.inference(t, noise, diffusion_steps=steps, embedding_scale=1))
     return (24000, np.concatenate(audios))
 
-
 with gr.Blocks() as vctk:
     with gr.Row():
         with gr.Column(scale=1):
             inp = gr.Textbox(label="Text", info="What would you like StyleTTS 2 to read? It works better on full sentences.", interactive=True)
             voice = gr.Dropdown(voicelist, label="Voice", info="Select a default voice.", value='m-us-2', interactive=True)
             multispeakersteps = gr.Slider(minimum=3, maximum=15, value=3, step=1, label="Diffusion Steps", info="Theoretically, higher should be better quality but slower, but we cannot notice a difference. Try with lower steps first - it is faster", interactive=True)
-            # use_gruut = gr.Checkbox(label="Use alternate phonemizer (Gruut) - Experimental")
         with gr.Column(scale=1):
             btn = gr.Button("Synthesize", variant="primary")
             audio = gr.Audio(interactive=False, label="Synthesized Audio", show_download_button=True, waveform_options={'waveform_progress_color': '#3C82F6'})
             btn.click(synthesize, inputs=[inp, voice, multispeakersteps], outputs=[audio], concurrency_limit=4)
+
 with gr.Blocks() as clone:
     with gr.Row():
         with gr.Column(scale=1):
-            clinp = gr.Textbox(label="Text", info="What would you like StyleTTS 2 to read? It works better on full sentences.", interactive=True)
-            clvoice = gr.Audio(label="Voice", interactive=True, type='filepath', max_length=300, waveform_options={'waveform_progress_color': '#3C82F6'})
-            vcsteps = gr.Slider(minimum=3, maximum=20, value=20, step=1, label="Diffusion Steps", info="Theoretically, higher should be better quality but slower, but we cannot notice a difference. Try with lower steps first - it is faster", interactive=True)
-            embscale = gr.Slider(minimum=1, maximum=10, value=1, step=0.1, label="Embedding Scale (READ WARNING BELOW)", info="Defaults to 1. WARNING: If you set this too high and generate text that's too short you will get static!", interactive=True)
+            clinp = gr.Textbox(label="Text", info="What would you like StyleTTS 2 to read?", interactive=True)
+
+            # ❌ BẢN GỐC:
+            # clvoice = gr.Audio(label="Voice", interactive=True, type='filepath', max_length=300, waveform_options={...})
+
+            # ✅ ĐÃ SỬA — XOÁ GIỚI HẠN max_length
+            clvoice = gr.Audio(
+                label="Voice",
+                interactive=True,
+                type='filepath',
+                waveform_options={'waveform_progress_color': '#3C82F6'}
+            )
+
+            vcsteps = gr.Slider(minimum=3, maximum=20, value=20, step=1, label="Diffusion Steps", info="Higher = better but slower", interactive=True)
+            embscale = gr.Slider(minimum=1, maximum=10, value=1, step=0.1, label="Embedding Scale", info="Default 1", interactive=True)
             alpha = gr.Slider(minimum=0, maximum=1, value=0.3, step=0.1, label="Alpha", info="Defaults to 0.3", interactive=True)
             beta = gr.Slider(minimum=0, maximum=1, value=0.7, step=0.1, label="Beta", info="Defaults to 0.7", interactive=True)
         with gr.Column(scale=1):
             clbtn = gr.Button("Synthesize", variant="primary")
             claudio = gr.Audio(interactive=False, label="Synthesized Audio", show_download_button=True, waveform_options={'waveform_progress_color': '#3C82F6'})
             clbtn.click(rn_clsynthesize, inputs=[clinp, clvoice, vcsteps, embscale, alpha, beta], outputs=[claudio], concurrency_limit=4)
-# with gr.Blocks() as longText:
-#     with gr.Row():
-#         with gr.Column(scale=1):
-#             lnginp = gr.Textbox(label="Text", info="What would you like StyleTTS 2 to read? It works better on full sentences.", interactive=True)
-#             lngvoice = gr.Dropdown(voicelist, label="Voice", info="Select a default voice.", value='m-us-1', interactive=True)
-#             lngsteps = gr.Slider(minimum=5, maximum=25, value=10, step=1, label="Diffusion Steps", info="Higher = better quality, but slower", interactive=True)
-#             lngpwd = gr.Textbox(label="Access code", info="This feature is in beta. You need an access code to use it as it uses more resources and we would like to prevent abuse")
-#         with gr.Column(scale=1):
-#             lngbtn = gr.Button("Synthesize", variant="primary")
-#             lngaudio = gr.Audio(interactive=False, label="Synthesized Audio")
-#             lngbtn.click(longsynthesize, inputs=[lnginp, lngvoice, lngsteps, lngpwd], outputs=[lngaudio], concurrency_limit=4)
+
 with gr.Blocks() as lj:
     with gr.Row():
         with gr.Column(scale=1):
-            ljinp = gr.Textbox(label="Text", info="What would you like StyleTTS 2 to read? It works better on full sentences.", interactive=True)
-            ljsteps = gr.Slider(minimum=3, maximum=20, value=3, step=1, label="Diffusion Steps", info="Theoretically, higher should be better quality but slower, but we cannot notice a difference. Try with lower steps first - it is faster", interactive=True)
+            ljinp = gr.Textbox(label="Text", info="What would you like StyleTTS 2 to read?", interactive=True)
+            ljsteps = gr.Slider(minimum=3, maximum=20, value=3, step=1, label="Diffusion Steps", interactive=True)
         with gr.Column(scale=1):
             ljbtn = gr.Button("Synthesize", variant="primary")
             ljaudio = gr.Audio(interactive=False, label="Synthesized Audio", waveform_options={'waveform_progress_color': '#3C82F6'})
             ljbtn.click(rn_ljsynthesize, inputs=[ljinp, ljsteps], outputs=[ljaudio], concurrency_limit=4)
+
 with gr.Blocks(title="StyleTTS 2", css="footer{display:none !important}", theme=theme) as demo:
     gr.Markdown(INTROTXT)
     gr.DuplicateButton("Duplicate Space")
-    # gr.TabbedInterface([vctk, clone, lj, longText], ['Multi-Voice', 'Voice Cloning', 'LJSpeech', 'Long Text [Beta]'])
     gr.TabbedInterface([vctk, clone, lj], ['Multi-Voice', 'Voice Cloning', 'LJSpeech', 'Long Text [Beta]'])
     gr.Markdown("""
-Demo by [mrfakename](https://twitter.com/realmrfakename). I am not affiliated with the StyleTTS 2 authors.
+Demo by mrfakename.
 
 Run this demo locally using Docker: