Spaces:
Runtime error
Runtime error
Upload app.py
Browse files
app.py
CHANGED
|
@@ -1,13 +1,5 @@
|
|
| 1 |
-
import os,re
|
| 2 |
-
|
| 3 |
-
logging.getLogger("urllib3").setLevel(logging.ERROR)
|
| 4 |
-
logging.getLogger("httpcore").setLevel(logging.ERROR)
|
| 5 |
-
logging.getLogger("httpx").setLevel(logging.ERROR)
|
| 6 |
-
logging.getLogger("asyncio").setLevel(logging.ERROR)
|
| 7 |
-
|
| 8 |
-
logging.getLogger("charset_normalizer").setLevel(logging.ERROR)
|
| 9 |
-
logging.getLogger("torchaudio._extension").setLevel(logging.ERROR)
|
| 10 |
-
import pdb
|
| 11 |
|
| 12 |
gpt_path = os.environ.get(
|
| 13 |
"gpt_path", "models/Taffy/Taffy-e5.ckpt"
|
|
@@ -57,6 +49,7 @@ else:
|
|
| 57 |
bert_model = bert_model.to(device)
|
| 58 |
|
| 59 |
|
|
|
|
| 60 |
def get_bert_feature(text, word2ph):
|
| 61 |
with torch.no_grad():
|
| 62 |
inputs = tokenizer(text, return_tensors="pt")
|
|
@@ -70,8 +63,15 @@ def get_bert_feature(text, word2ph):
|
|
| 70 |
repeat_feature = res[i].repeat(word2ph[i], 1)
|
| 71 |
phone_level_feature.append(repeat_feature)
|
| 72 |
phone_level_feature = torch.cat(phone_level_feature, dim=0)
|
|
|
|
| 73 |
return phone_level_feature.T
|
| 74 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 75 |
class DictToAttrRecursive(dict):
|
| 76 |
def __init__(self, input_dict):
|
| 77 |
super().__init__(input_dict)
|
|
@@ -99,6 +99,12 @@ class DictToAttrRecursive(dict):
|
|
| 99 |
except KeyError:
|
| 100 |
raise AttributeError(f"Attribute {item} not found")
|
| 101 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 102 |
ssl_model = cnhubert.get_model()
|
| 103 |
if is_half == True:
|
| 104 |
ssl_model = ssl_model.half().to(device)
|
|
@@ -117,8 +123,7 @@ def change_sovits_weights(sovits_path):
|
|
| 117 |
n_speakers=hps.data.n_speakers,
|
| 118 |
**hps.model
|
| 119 |
)
|
| 120 |
-
|
| 121 |
-
del vq_model.enc_q
|
| 122 |
if is_half == True:
|
| 123 |
vq_model = vq_model.half().to(device)
|
| 124 |
else:
|
|
@@ -160,88 +165,14 @@ def get_spepc(hps, filename):
|
|
| 160 |
return spec
|
| 161 |
|
| 162 |
|
| 163 |
-
dict_language={
|
| 164 |
-
|
| 165 |
-
|
| 166 |
-
|
| 167 |
-
|
| 168 |
-
|
| 169 |
-
|
| 170 |
-
|
| 171 |
-
pattern = re.compile(r'[a-zA-Z. ]+')
|
| 172 |
-
textlist = []
|
| 173 |
-
langlist = []
|
| 174 |
-
pos = 0
|
| 175 |
-
for match in pattern.finditer(sentence):
|
| 176 |
-
start, end = match.span()
|
| 177 |
-
if start > pos:
|
| 178 |
-
textlist.append(sentence[pos:start])
|
| 179 |
-
langlist.append(language)
|
| 180 |
-
textlist.append(sentence[start:end])
|
| 181 |
-
langlist.append("en")
|
| 182 |
-
pos = end
|
| 183 |
-
if pos < len(sentence):
|
| 184 |
-
textlist.append(sentence[pos:])
|
| 185 |
-
langlist.append(language)
|
| 186 |
-
|
| 187 |
-
return textlist, langlist
|
| 188 |
-
|
| 189 |
-
|
| 190 |
-
def clean_text_inf(text, language):
|
| 191 |
-
phones, word2ph, norm_text = clean_text(text, language)
|
| 192 |
-
phones = cleaned_text_to_sequence(phones)
|
| 193 |
-
|
| 194 |
-
return phones, word2ph, norm_text
|
| 195 |
-
def get_bert_inf(phones, word2ph, norm_text, language):
|
| 196 |
-
if language == "zh":
|
| 197 |
-
bert = get_bert_feature(norm_text, word2ph).to(device)
|
| 198 |
-
else:
|
| 199 |
-
bert = torch.zeros(
|
| 200 |
-
(1024, len(phones)),
|
| 201 |
-
dtype=torch.float16 if is_half == True else torch.float32,
|
| 202 |
-
).to(device)
|
| 203 |
-
|
| 204 |
-
return bert
|
| 205 |
-
|
| 206 |
-
|
| 207 |
-
def nonen_clean_text_inf(text, language):
|
| 208 |
-
textlist, langlist = splite_en_inf(text, language)
|
| 209 |
-
phones_list = []
|
| 210 |
-
word2ph_list = []
|
| 211 |
-
norm_text_list = []
|
| 212 |
-
for i in range(len(textlist)):
|
| 213 |
-
lang = langlist[i]
|
| 214 |
-
phones, word2ph, norm_text = clean_text_inf(textlist[i], lang)
|
| 215 |
-
phones_list.append(phones)
|
| 216 |
-
if lang == "en" or "ja":
|
| 217 |
-
pass
|
| 218 |
-
else:
|
| 219 |
-
word2ph_list.append(word2ph)
|
| 220 |
-
norm_text_list.append(norm_text)
|
| 221 |
-
print(word2ph_list)
|
| 222 |
-
phones = sum(phones_list, [])
|
| 223 |
-
word2ph = sum(word2ph_list, [])
|
| 224 |
-
norm_text = ' '.join(norm_text_list)
|
| 225 |
-
|
| 226 |
-
return phones, word2ph, norm_text
|
| 227 |
-
|
| 228 |
-
|
| 229 |
-
def nonen_get_bert_inf(text, language):
|
| 230 |
-
textlist, langlist = splite_en_inf(text, language)
|
| 231 |
-
print(textlist)
|
| 232 |
-
print(langlist)
|
| 233 |
-
bert_list = []
|
| 234 |
-
for i in range(len(textlist)):
|
| 235 |
-
text = textlist[i]
|
| 236 |
-
lang = langlist[i]
|
| 237 |
-
phones, word2ph, norm_text = clean_text_inf(text, lang)
|
| 238 |
-
bert = get_bert_inf(phones, word2ph, norm_text, lang)
|
| 239 |
-
bert_list.append(bert)
|
| 240 |
-
bert = torch.cat(bert_list, dim=1)
|
| 241 |
-
|
| 242 |
-
return bert
|
| 243 |
-
|
| 244 |
-
def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,how_to_cut=("不切")):
|
| 245 |
t0 = ttime()
|
| 246 |
prompt_text = prompt_text.strip("\n")
|
| 247 |
prompt_language, text = prompt_language, text.strip("\n")
|
|
@@ -270,37 +201,28 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
|
|
| 270 |
t1 = ttime()
|
| 271 |
prompt_language = dict_language[prompt_language]
|
| 272 |
text_language = dict_language[text_language]
|
| 273 |
-
|
| 274 |
-
|
| 275 |
-
|
| 276 |
-
else:
|
| 277 |
-
phones1, word2ph1, norm_text1 = nonen_clean_text_inf(prompt_text, prompt_language)
|
| 278 |
-
if(how_to_cut==("凑五句一切")):text=cut1(text)
|
| 279 |
-
elif(how_to_cut==("凑50字一切")):text=cut2(text)
|
| 280 |
-
elif(how_to_cut==("按中文句号。切")):text=cut3(text)
|
| 281 |
-
elif(how_to_cut==("按英文句号.切")):text=cut4(text)
|
| 282 |
-
text = text.replace("\n\n","\n").replace("\n\n","\n").replace("\n\n","\n")
|
| 283 |
-
if(text[-1]not in splits):text+="。"if text_language!="en"else "."
|
| 284 |
-
texts=text.split("\n")
|
| 285 |
audio_opt = []
|
| 286 |
-
if prompt_language == "en":
|
| 287 |
-
bert1 = get_bert_inf(phones1, word2ph1, norm_text1, prompt_language)
|
| 288 |
-
else:
|
| 289 |
-
bert1 = nonen_get_bert_inf(prompt_text, prompt_language)
|
| 290 |
|
| 291 |
for text in texts:
|
| 292 |
# 解决输入目标文本的空行导致报错的问题
|
| 293 |
if (len(text.strip()) == 0):
|
| 294 |
continue
|
| 295 |
-
|
| 296 |
-
|
|
|
|
|
|
|
| 297 |
else:
|
| 298 |
-
|
| 299 |
-
|
| 300 |
-
|
| 301 |
-
|
|
|
|
|
|
|
| 302 |
else:
|
| 303 |
-
bert2 =
|
| 304 |
bert = torch.cat([bert1, bert2], 1)
|
| 305 |
|
| 306 |
all_phoneme_ids = torch.LongTensor(phones1 + phones2).to(device).unsqueeze(0)
|
|
@@ -458,7 +380,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
|
|
| 458 |
### <center>⚠️在线端不稳定且生成速度较慢,强烈建议下载模型本地推理!\n
|
| 459 |
""")
|
| 460 |
# with gr.Tabs():
|
| 461 |
-
|
| 462 |
with gr.Group():
|
| 463 |
gr.Markdown(value="*参考音频选择(必选)")
|
| 464 |
with gr.Row():
|
|
|
|
| 1 |
+
import os,re
|
| 2 |
+
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
|
| 4 |
gpt_path = os.environ.get(
|
| 5 |
"gpt_path", "models/Taffy/Taffy-e5.ckpt"
|
|
|
|
| 49 |
bert_model = bert_model.to(device)
|
| 50 |
|
| 51 |
|
| 52 |
+
# bert_model=bert_model.to(device)
|
| 53 |
def get_bert_feature(text, word2ph):
|
| 54 |
with torch.no_grad():
|
| 55 |
inputs = tokenizer(text, return_tensors="pt")
|
|
|
|
| 63 |
repeat_feature = res[i].repeat(word2ph[i], 1)
|
| 64 |
phone_level_feature.append(repeat_feature)
|
| 65 |
phone_level_feature = torch.cat(phone_level_feature, dim=0)
|
| 66 |
+
# if(is_half==True):phone_level_feature=phone_level_feature.half()
|
| 67 |
return phone_level_feature.T
|
| 68 |
|
| 69 |
+
|
| 70 |
+
n_semantic = 1024
|
| 71 |
+
|
| 72 |
+
dict_s2=torch.load(sovits_path,map_location="cpu")
|
| 73 |
+
hps=dict_s2["config"]
|
| 74 |
+
|
| 75 |
class DictToAttrRecursive(dict):
|
| 76 |
def __init__(self, input_dict):
|
| 77 |
super().__init__(input_dict)
|
|
|
|
| 99 |
except KeyError:
|
| 100 |
raise AttributeError(f"Attribute {item} not found")
|
| 101 |
|
| 102 |
+
|
| 103 |
+
hps = DictToAttrRecursive(hps)
|
| 104 |
+
|
| 105 |
+
hps.model.semantic_frame_rate = "25hz"
|
| 106 |
+
dict_s1 = torch.load(gpt_path, map_location="cpu")
|
| 107 |
+
config = dict_s1["config"]
|
| 108 |
ssl_model = cnhubert.get_model()
|
| 109 |
if is_half == True:
|
| 110 |
ssl_model = ssl_model.half().to(device)
|
|
|
|
| 123 |
n_speakers=hps.data.n_speakers,
|
| 124 |
**hps.model
|
| 125 |
)
|
| 126 |
+
del vq_model.enc_q
|
|
|
|
| 127 |
if is_half == True:
|
| 128 |
vq_model = vq_model.half().to(device)
|
| 129 |
else:
|
|
|
|
| 165 |
return spec
|
| 166 |
|
| 167 |
|
| 168 |
+
dict_language = {"中文": "zh", "英文": "en", "日文": "ja"}
|
| 169 |
+
|
| 170 |
+
|
| 171 |
+
def get_tts_wav(selected_text, prompt_text, prompt_language, text, text_language):
|
| 172 |
+
ref_wav_path = text_to_audio_mappings.get(selected_text, "")
|
| 173 |
+
if not ref_wav_path:
|
| 174 |
+
print("Audio file not found for the selected text.")
|
| 175 |
+
return
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 176 |
t0 = ttime()
|
| 177 |
prompt_text = prompt_text.strip("\n")
|
| 178 |
prompt_language, text = prompt_language, text.strip("\n")
|
|
|
|
| 201 |
t1 = ttime()
|
| 202 |
prompt_language = dict_language[prompt_language]
|
| 203 |
text_language = dict_language[text_language]
|
| 204 |
+
phones1, word2ph1, norm_text1 = clean_text(prompt_text, prompt_language)
|
| 205 |
+
phones1 = cleaned_text_to_sequence(phones1)
|
| 206 |
+
texts = text.split("\n")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 207 |
audio_opt = []
|
|
|
|
|
|
|
|
|
|
|
|
|
| 208 |
|
| 209 |
for text in texts:
|
| 210 |
# 解决输入目标文本的空行导致报错的问题
|
| 211 |
if (len(text.strip()) == 0):
|
| 212 |
continue
|
| 213 |
+
phones2, word2ph2, norm_text2 = clean_text(text, text_language)
|
| 214 |
+
phones2 = cleaned_text_to_sequence(phones2)
|
| 215 |
+
if prompt_language == "zh":
|
| 216 |
+
bert1 = get_bert_feature(norm_text1, word2ph1).to(device)
|
| 217 |
else:
|
| 218 |
+
bert1 = torch.zeros(
|
| 219 |
+
(1024, len(phones1)),
|
| 220 |
+
dtype=torch.float16 if is_half == True else torch.float32,
|
| 221 |
+
).to(device)
|
| 222 |
+
if text_language == "zh":
|
| 223 |
+
bert2 = get_bert_feature(norm_text2, word2ph2).to(device)
|
| 224 |
else:
|
| 225 |
+
bert2 = torch.zeros((1024, len(phones2))).to(bert1)
|
| 226 |
bert = torch.cat([bert1, bert2], 1)
|
| 227 |
|
| 228 |
all_phoneme_ids = torch.LongTensor(phones1 + phones2).to(device).unsqueeze(0)
|
|
|
|
| 380 |
### <center>⚠️在线端不稳定且生成速度较慢,强烈建议下载模型本地推理!\n
|
| 381 |
""")
|
| 382 |
# with gr.Tabs():
|
| 383 |
+
# with gr.TabItem(i18n("伴奏人声分离&去混响&去回声")):
|
| 384 |
with gr.Group():
|
| 385 |
gr.Markdown(value="*参考音频选择(必选)")
|
| 386 |
with gr.Row():
|