Update app.py

app.py

@@ -7,7 +7,7 @@ import torchaudio
 import gradio as gr
 import re
 
-llasa_model_id = 'OmniAICreator/Galgame-Llasa-1B-v2'
+llasa_model_id = 'OmniAICreator/Galgame-Llasa-1B-v3'
 
 tokenizer = AutoTokenizer.from_pretrained(llasa_model_id)
 
@@ -78,17 +78,12 @@ def normalize(text: str) -> str:
     text = text.translate(FULLWIDTH_DIGITS_TO_HALFWIDTH)
     text = text.translate(HALFWIDTH_KATAKANA_TO_FULLWIDTH)
 
-    text = re.sub(r"…{2,}", "…", text)
-    text = re.sub(r"ー{2,}", "ー", text)
+    text = re.sub(r"…{3,}", "……", text)
 
     def replace_special_chars(match):
         seq = match.group(0)
         return seq[0] if len(set(seq)) == 1 else seq[0] + seq[-1]
 
-    text = re.sub(r"[!?♪♡]{2,}", replace_special_chars, text)
-
-    repeated = "ッっあいうえおんぁぃぅぇぉゃゅょアイウエオンァィゥェォャュョ"
-    text = re.sub(f"([{repeated}])\\1{{2,}}", r"\1\1", text)
     return text
 
 def ids_to_speech_tokens(speech_ids):
@@ -216,7 +211,7 @@ def infer(sample_audio_path, target_text, temperature, top_p, repetition_penalty
         return (16000, gen_wav[0, 0, :].cpu().numpy())
 
 with gr.Blocks() as app_tts:
-    gr.Markdown("# Galgame Llasa 1B v2")
+    gr.Markdown("# Galgame Llasa 1B v3")
     ref_audio_input = gr.Audio(label="Reference Audio", type="filepath")
     gen_text_input = gr.Textbox(label="Text to Generate", lines=10)