Update inference_webui.py

GPT_SoVITS/inference_webui.py

@@ -1,4 +1,5 @@
 import os, re, logging
+
 logging.getLogger("markdown_it").setLevel(logging.ERROR)
 logging.getLogger("urllib3").setLevel(logging.ERROR)
 logging.getLogger("httpcore").setLevel(logging.ERROR)
@@ -46,6 +47,7 @@ from transformers import AutoModelForMaskedLM, AutoTokenizer
 import numpy as np
 import librosa, torch
 from feature_extractor import cnhubert
+
 cnhubert.cnhubert_base_path = cnhubert_base_path
 
 from module.models import SynthesizerTrn
@@ -56,6 +58,7 @@ from time import time as ttime
 from module.mel_processing import spectrogram_torch
 from my_utils import load_audio
 from tools.i18n.i18n import I18nAuto
+
 i18n = I18nAuto()
 
 os.environ['PYTORCH_ENABLE_MPS_FALLBACK'] = '1'  # 确保直接启动推理UI时也能够设置。
@@ -74,6 +77,7 @@ if is_half == True:
 else:
     bert_model = bert_model.to(device)
 
+
 def get_bert_feature(text, word2ph):
     with torch.no_grad():
         inputs = tokenizer(text, return_tensors="pt")
@@ -89,6 +93,7 @@ def get_bert_feature(text, word2ph):
     phone_level_feature = torch.cat(phone_level_feature, dim=0)
     return phone_level_feature.T
 
+
 class DictToAttrRecursive(dict):
     def __init__(self, input_dict):
         super().__init__(input_dict)
@@ -123,6 +128,7 @@ if is_half == True:
 else:
     ssl_model = ssl_model.to(device)
 
+
 def change_sovits_weights(sovits_path):
     global vq_model, hps
     dict_s2 = torch.load(sovits_path, map_location="cpu")
@@ -143,9 +149,13 @@ def change_sovits_weights(sovits_path):
         vq_model = vq_model.to(device)
     vq_model.eval()
     print(vq_model.load_state_dict(dict_s2["weight"], strict=False))
-    with open("./sweight.txt","w",encoding="utf-8")as f:f.write(sovits_path)
+    with open("./sweight.txt", "w", encoding="utf-8") as f:
+        f.write(sovits_path)
+
+
 change_sovits_weights(sovits_path)
 
+
 def change_gpt_weights(gpt_path):
     global hz, max_sec, t2s_model, config
     hz = 50
@@ -161,8 +171,11 @@ def change_gpt_weights(gpt_path):
     total = sum([param.nelement() for param in t2s_model.parameters()])
     print("Number of parameter: %.2fM" % (total / 1e6))
     with open("./gweight.txt", "w", encoding="utf-8") as f: f.write(gpt_path)
+
+
 change_gpt_weights(gpt_path)
 
+
 def get_spepc(hps, filename):
     audio = load_audio(filename, int(hps.data.sampling_rate))
     audio = torch.FloatTensor(audio)
@@ -262,12 +275,16 @@ def nonen_get_bert_inf(text, language):
 
     return bert
 
+
 splits = {"，", "。", "？", "！", ",", ".", "?", "!", "~", ":", "：", "—", "…", }
+
+
 def get_first(text):
     pattern = "[" + "".join(re.escape(sep) for sep in splits) + "]"
     text = re.split(pattern, text)[0].strip()
     return text
 
+
 def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language, how_to_cut=i18n("不切")):
     t0 = ttime()
     prompt_text = prompt_text.strip("\n")
@@ -307,10 +324,16 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
         phones1, word2ph1, norm_text1 = clean_text_inf(prompt_text, prompt_language)
     else:
         phones1, word2ph1, norm_text1 = nonen_clean_text_inf(prompt_text, prompt_language)
-    if(how_to_cut==i18n("凑四句一切")):text=cut1(text)
+    if (how_to_cut == i18n("凑四句一切")):
-    elif(how_to_cut==i18n("凑50字一切")):text=cut2(text)
+        text = cut1(text)
-    elif(how_to_cut==i18n("按中文句号。切")):text=cut3(text)
+    elif (how_to_cut == i18n("凑50字一切")):
-    elif(how_to_cut==i18n("按英文句号.切")):text=cut4(text)
+        text = cut2(text)
+    elif (how_to_cut == i18n("按中文句号。切")):
+        text = cut3(text)
+    elif (how_to_cut == i18n("按英文句号.切")):
+        text = cut4(text)
+    elif (how_to_cut == i18n("按标点符号切")):
+        text = cut5(text)
     text = text.replace("\n\n", "\n").replace("\n\n", "\n").replace("\n\n", "\n")
     print(i18n("实际输入的目标文本(切句后):"), text)
     texts = text.split("\n")
@@ -380,6 +403,7 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
         np.int16
     )
 
+
 def split(todo_text):
     todo_text = todo_text.replace("……", "。").replace("——", "，")
     if todo_text[-1] not in splits:
@@ -440,10 +464,25 @@ def cut2(inp):
 def cut3(inp):
     inp = inp.strip("\n")
     return "\n".join(["%s" % item for item in inp.strip("。").split("。")])
+
+
 def cut4(inp):
     inp = inp.strip("\n")
     return "\n".join(["%s" % item for item in inp.strip(".").split(".")])
 
+
+# contributed by https://github.com/AI-Hobbyist/GPT-SoVITS/blob/main/GPT_SoVITS/inference_webui.py
+def cut5(inp):
+    # if not re.search(r'[^\w\s]', inp[-1]):
+    # inp += '。'
+    inp = inp.strip("\n")
+    punds = r'[,.;?!、，。？！;：]'
+    items = re.split(f'({punds})', inp)
+    items = ["".join(group) for group in zip(items[::2], items[1::2])]
+    opt = "\n".join(items)
+    return opt
+
+
 def custom_sort_key(s):
     # 使用正则表达式提取字符串中的数字部分和非数字部分
     parts = re.split('(\d+)', s)
@@ -451,16 +490,20 @@ def custom_sort_key(s):
     parts = [int(part) if part.isdigit() else part for part in parts]
     return parts
 
+
 def change_choices():
     SoVITS_names, GPT_names = get_weights_names()
     return {"choices": sorted(SoVITS_names, key=custom_sort_key), "__type__": "update"}, {"choices": sorted(GPT_names, key=custom_sort_key), "__type__": "update"}
 
+
 pretrained_sovits_name = "GPT_SoVITS/pretrained_models/s2G488k.pth"
 pretrained_gpt_name = "GPT_SoVITS/pretrained_models/s1bert25hz-2kh-longer-epoch=68e-step=50232.ckpt"
 SoVITS_weight_root = "SoVITS_weights"
 GPT_weight_root = "GPT_weights"
 os.makedirs(SoVITS_weight_root, exist_ok=True)
 os.makedirs(GPT_weight_root, exist_ok=True)
+
+
 def get_weights_names():
     SoVITS_names = [pretrained_sovits_name]
     for name in os.listdir(SoVITS_weight_root):
@@ -469,6 +512,8 @@ def get_weights_names():
     for name in os.listdir(GPT_weight_root):
         if name.endswith(".ckpt"): GPT_names.append("%s/%s" % (GPT_weight_root, name))
     return SoVITS_names, GPT_names
+
+
 SoVITS_names, GPT_names = get_weights_names()
 
 with gr.Blocks(title="GPT-SoVITS WebUI") as app:
@@ -499,8 +544,8 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
             )
             how_to_cut = gr.Radio(
                 label=i18n("怎么切"),
-                choices=[i18n("不切"),i18n("凑四句一切"),i18n("凑50字一切"),i18n("按中文句号。切"),i18n("按英文句号.切"),],
+                choices=[i18n("不切"), i18n("凑四句一切"), i18n("凑50字一切"), i18n("按中文句号。切"), i18n("按英文句号.切"), i18n("按标点符号切"), ],
-                value=i18n("凑50字一切"),
+                value=i18n("凑四句一切"),
                 interactive=True,
             )
             inference_button = gr.Button(i18n("合成语音"), variant="primary")
@@ -519,11 +564,13 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
             button2 = gr.Button(i18n("凑50字一切"), variant="primary")
             button3 = gr.Button(i18n("按中文句号。切"), variant="primary")
             button4 = gr.Button(i18n("按英文句号.切"), variant="primary")
+            button5 = gr.Button(i18n("按标点符号切"), variant="primary")
             text_opt = gr.Textbox(label=i18n("切分后文本"), value="")
             button1.click(cut1, [text_inp], [text_opt])
             button2.click(cut2, [text_inp], [text_opt])
             button3.click(cut3, [text_inp], [text_opt])
             button4.click(cut4, [text_inp], [text_opt])
+            button5.click(cut5, [text_inp], [text_opt])
         gr.Markdown(value=i18n("后续将支持混合语种编码文本输入。"))
 
 app.queue(concurrency_count=511, max_size=1022).launch(