Update app.py

app.py

@@ -96,7 +96,7 @@ def clsynthesize(text, voice, vcsteps, embscale, alpha, beta, progress=gr.Progre
         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 ljsynthesize(text, steps, progress=gr.Progress()):
+def ljsynthesize(text, steps,embscale, progress=gr.Progress()):
     # if text.strip() == "":
     #     raise gr.Error("You must enter some text")
     # # if global_phonemizer.phonemize([text]) > 300:
@@ -114,7 +114,7 @@ def ljsynthesize(text, steps, progress=gr.Progress()):
     texts = txtsplit(text)
     audios = []
     for t in progress.tqdm(texts):
-        audios.append(ljspeechimportable.inference(t, noise, diffusion_steps=steps, embedding_scale=1))
+        audios.append(ljspeechimportable.inference(t, noise, diffusion_steps=steps, embedding_scale=embscale))
     return (24000, np.concatenate(audios))
 
 
@@ -157,6 +157,7 @@ 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)
+            embscale = gr.Slider(minimum=1, maximum=10, value=1, step=0.1, label="Embedding Scale (READ WARNING BELOW)", info="Defaults to 1. これを上げたらパフォーマンスがもっとエモーショナルになる、増やしすぎるとだめになるので、ご注意ください", 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)
         with gr.Column(scale=1):
             ljbtn = gr.Button("Synthesize", variant="primary")