feat: update import path

Signed-off-by: Guoxian Dai <guoxian@peta-mini1.lan>

.gitignore

@@ -13,6 +13,6 @@ SoVITS_weights
 GPT_weights_v2
 SoVITS_weights_v2
 TEMP
-weight.json
+#weight.json
 ffmpeg*
 ffprobe*

GPT_SoVITS/AR/models/t2s_lightning_module.py

@@ -8,9 +8,9 @@ from typing import Dict
 
 import torch
 from pytorch_lightning import LightningModule
-from AR.models.t2s_model import Text2SemanticDecoder
-from AR.modules.lr_schedulers import WarmupCosineLRSchedule
-from AR.modules.optim import ScaledAdam
+from GPT_SoVITS.AR.models.t2s_model import Text2SemanticDecoder
+from GPT_SoVITS.AR.modules.lr_schedulers import WarmupCosineLRSchedule
+from GPT_SoVITS.AR.modules.optim import ScaledAdam
 
 class Text2SemanticLightningModule(LightningModule):
     def __init__(self, config, output_dir, is_train=True):

GPT_SoVITS/AR/models/t2s_model.py

@@ -5,8 +5,8 @@ from typing import List, Optional
 import torch
 from tqdm import tqdm
 
-from AR.models.utils import make_pad_mask
-from AR.models.utils import (
+from GPT_SoVITS.AR.models.utils import make_pad_mask
+from GPT_SoVITS.AR.models.utils import (
     topk_sampling,
     sample,
     logits_to_probs,
@@ -15,11 +15,11 @@ from AR.models.utils import (
     make_reject_y,
     get_batch_logps
 )
-from AR.modules.embedding import SinePositionalEmbedding
-from AR.modules.embedding import TokenEmbedding
-from AR.modules.transformer import LayerNorm
-from AR.modules.transformer import TransformerEncoder
-from AR.modules.transformer import TransformerEncoderLayer
+from GPT_SoVITS.AR.modules.embedding import SinePositionalEmbedding
+from GPT_SoVITS.AR.modules.embedding import TokenEmbedding
+from GPT_SoVITS.AR.modules.transformer import LayerNorm
+from GPT_SoVITS.AR.modules.transformer import TransformerEncoder
+from GPT_SoVITS.AR.modules.transformer import TransformerEncoderLayer
 from torch import nn
 from torch.nn import functional as F
 from torchmetrics.classification import MulticlassAccuracy

GPT_SoVITS/AR/modules/activation.py

@@ -12,7 +12,7 @@ from torch.nn.modules.linear import NonDynamicallyQuantizableLinear
 from torch.nn.parameter import Parameter
 
 from torch.nn import functional as F
-from AR.modules.patched_mha_with_cache import multi_head_attention_forward_patched
+from GPT_SoVITS.AR.modules.patched_mha_with_cache import multi_head_attention_forward_patched
 
 F.multi_head_attention_forward = multi_head_attention_forward_patched
 

GPT_SoVITS/AR/modules/transformer.py

@@ -10,8 +10,8 @@ from typing import Tuple
 from typing import Union
 
 import torch
-from AR.modules.activation import MultiheadAttention
-from AR.modules.scaling import BalancedDoubleSwish
+from GPT_SoVITS.AR.modules.activation import MultiheadAttention
+from GPT_SoVITS.AR.modules.scaling import BalancedDoubleSwish
 from torch import nn
 from torch import Tensor
 from torch.nn import functional as F

GPT_SoVITS/TTS_infer_pack/TTS.py

@@ -21,8 +21,8 @@ from feature_extractor.cnhubert import CNHubert
 from module.models import SynthesizerTrn
 import librosa
 from time import time as ttime
-from tools.i18n.i18n import I18nAuto, scan_language_list
-from tools.my_utils import load_audio
+from GPT_SoVITS.tools.i18n.i18n import I18nAuto, scan_language_list
+from GPT_SoVITS.tools.my_utils import load_audio
 from module.mel_processing import spectrogram_torch
 from TTS_infer_pack.text_segmentation_method import splits
 from TTS_infer_pack.TextPreprocessor import TextPreprocessor

GPT_SoVITS/TTS_infer_pack/TextPreprocessor.py

@@ -15,7 +15,7 @@ from text import cleaned_text_to_sequence
 from transformers import AutoModelForMaskedLM, AutoTokenizer
 from TTS_infer_pack.text_segmentation_method import split_big_text, splits, get_method as get_seg_method
 
-from tools.i18n.i18n import I18nAuto, scan_language_list
+from GPT_SoVITS.tools.i18n.i18n import I18nAuto, scan_language_list
 
 language=os.environ.get("language","Auto")
 language=sys.argv[-1] if sys.argv[-1] in scan_language_list() else language

GPT_SoVITS/feature_extractor/cnhubert.py

@@ -16,7 +16,7 @@ from transformers import (
     HubertModel,
 )
 
-import utils
+import GPT_SoVITS.utils
 import torch.nn as nn
 
 cnhubert_base_path = None
@@ -102,9 +102,10 @@ def get_content(hmodel, wav_16k_tensor):
 
 
 if __name__ == "__main__":
+    from GPT_SoVITS.utils import load_wav_to_torch_and_resample
     model = get_model()
     src_path = "/Users/Shared/原音频2.wav"
-    wav_16k_tensor = utils.load_wav_to_torch_and_resample(src_path, 16000)
+    wav_16k_tensor = load_wav_to_torch_and_resample(src_path, 16000)
     model = model
     wav_16k_tensor = wav_16k_tensor
     feats = get_content(model, wav_16k_tensor)

GPT_SoVITS/inference_cli.py

@@ -2,9 +2,15 @@ import argparse
 import os
 import soundfile as sf
 
-from tools.i18n.i18n import I18nAuto
+import GPT_SoVITS
+import sys
+
+from GPT_SoVITS.tools.i18n.i18n import I18nAuto
 from GPT_SoVITS.inference_webui import change_gpt_weights, change_sovits_weights, get_tts_wav
 
+
+
+
 i18n = I18nAuto()
 
 def synthesize(GPT_model_path, SoVITS_model_path, ref_audio_path, ref_text_path, ref_language, target_text_path, target_language, output_path):
@@ -18,6 +24,7 @@ def synthesize(GPT_model_path, SoVITS_model_path, ref_audio_path, ref_text_path,
 
     # Change model weights
     change_gpt_weights(gpt_path=GPT_model_path)
+
     change_sovits_weights(sovits_path=SoVITS_model_path)
 
     # Synthesize audio
@@ -29,6 +36,8 @@ def synthesize(GPT_model_path, SoVITS_model_path, ref_audio_path, ref_text_path,
     
     result_list = list(synthesis_result)
 
+    if not os.path.isdir(output_path):
+        os.makedirs(output_path, exist_ok=True)
     if result_list:
         last_sampling_rate, last_audio_data = result_list[-1]
         output_wav_path = os.path.join(output_path, "output.wav")

GPT_SoVITS/inference_gui.py

@@ -5,7 +5,7 @@ from PyQt5.QtWidgets import QApplication, QMainWindow, QLabel, QLineEdit, QPushB
 from PyQt5.QtWidgets import QGridLayout, QVBoxLayout, QWidget, QFileDialog, QStatusBar, QComboBox
 import soundfile as sf
 
-from tools.i18n.i18n import I18nAuto
+from GPT_SoVITS.tools.i18n.i18n import I18nAuto
 i18n = I18nAuto()
 
 from inference_webui import gpt_path, sovits_path, change_gpt_weights, change_sovits_weights, get_tts_wav

GPT_SoVITS/inference_webui.py

@@ -27,8 +27,8 @@ try:
 except:...
 
 version=os.environ.get("version","v2")
-pretrained_sovits_name=["GPT_SoVITS/pretrained_models/gsv-v2final-pretrained/s2G2333k.pth", "GPT_SoVITS/pretrained_models/s2G488k.pth"]
-pretrained_gpt_name=["GPT_SoVITS/pretrained_models/gsv-v2final-pretrained/s1bert25hz-5kh-longer-epoch=12-step=369668.ckpt", "GPT_SoVITS/pretrained_models/s1bert25hz-2kh-longer-epoch=68e-step=50232.ckpt"]
+pretrained_sovits_name=["pretrained_models/gsv-v2final-pretrained/s2G2333k.pth", "pretrained_models/s2G488k.pth"]
+pretrained_gpt_name=["pretrained_models/gsv-v2final-pretrained/s1bert25hz-5kh-longer-epoch=12-step=369668.ckpt", "pretrained_models/s1bert25hz-2kh-longer-epoch=68e-step=50232.ckpt"]
 
 _ =[[],[]]
 for i in range(2):
@@ -39,13 +39,18 @@ for i in range(2):
 pretrained_gpt_name,pretrained_sovits_name = _
     
 
+weight_json_file_path = os.path.join(
+    os.path.dirname(__file__),
+    "weight.json"
+)
 
-if os.path.exists(f"./weight.json"):
+print("weight_json_file_path: ", weight_json_file_path)
+if os.path.exists(weight_json_file_path):
     pass
 else:
-    with open(f"./weight.json", 'w', encoding="utf-8") as file:json.dump({'GPT':{},'SoVITS':{}},file)
+    with open(weight_json_file_path, 'w', encoding="utf-8") as file:json.dump({'GPT':{},'SoVITS':{}},file)
 
-with open(f"./weight.json", 'r', encoding="utf-8") as file:
+with open(weight_json_file_path, 'r', encoding="utf-8") as file:
     weight_data = file.read()
     weight_data=json.loads(weight_data)
     gpt_path = os.environ.get(
@@ -62,11 +67,12 @@ with open(f"./weight.json", 'r', encoding="utf-8") as file:
 # )
 # sovits_path = os.environ.get("sovits_path", pretrained_sovits_name)
 cnhubert_base_path = os.environ.get(
-    "cnhubert_base_path", "GPT_SoVITS/pretrained_models/chinese-hubert-base"
+    "cnhubert_base_path", "pretrained_models/chinese-hubert-base"
 )
 bert_path = os.environ.get(
-    "bert_path", "GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large"
+    "bert_path", "pretrained_models/chinese-roberta-wwm-ext-large"
 )
+
 infer_ttswebui = os.environ.get("infer_ttswebui", 9872)
 infer_ttswebui = int(infer_ttswebui)
 is_share = os.environ.get("is_share", "False")
@@ -79,18 +85,18 @@ import gradio as gr
 from transformers import AutoModelForMaskedLM, AutoTokenizer
 import numpy as np
 import librosa
-from feature_extractor import cnhubert
+from GPT_SoVITS.feature_extractor import cnhubert
 
 cnhubert.cnhubert_base_path = cnhubert_base_path
 
-from module.models import SynthesizerTrn
-from AR.models.t2s_lightning_module import Text2SemanticLightningModule
-from text import cleaned_text_to_sequence
-from text.cleaner import clean_text
+from GPT_SoVITS.module.models import SynthesizerTrn
+from GPT_SoVITS.AR.models.t2s_lightning_module import Text2SemanticLightningModule
+from GPT_SoVITS.text import cleaned_text_to_sequence
+from GPT_SoVITS.text.cleaner import clean_text
 from time import time as ttime
-from module.mel_processing import spectrogram_torch
-from tools.my_utils import load_audio
-from tools.i18n.i18n import I18nAuto, scan_language_list
+from GPT_SoVITS.module.mel_processing import spectrogram_torch
+from GPT_SoVITS.tools.my_utils import load_audio
+from GPT_SoVITS.tools.i18n.i18n import I18nAuto, scan_language_list
 
 language=os.environ.get("language","Auto")
 language=sys.argv[-1] if sys.argv[-1] in scan_language_list() else language
@@ -126,6 +132,7 @@ dict_language_v2 = {
 }
 dict_language = dict_language_v1 if version =='v1' else dict_language_v2
 
+
 tokenizer = AutoTokenizer.from_pretrained(bert_path)
 bert_model = AutoModelForMaskedLM.from_pretrained(bert_path)
 if is_half == True:
@@ -187,6 +194,9 @@ else:
 
 def change_sovits_weights(sovits_path,prompt_language=None,text_language=None):
     global vq_model, hps, version, dict_language
+    print("sovits_path: ", sovits_path)
+    print("os.path.isfile(sovits_path) = ", os.path.isfile(sovits_path))
+
     dict_s2 = torch.load(sovits_path, map_location="cpu")
     hps = dict_s2["config"]
     hps = DictToAttrRecursive(hps)
@@ -212,11 +222,11 @@ def change_sovits_weights(sovits_path,prompt_language=None,text_language=None):
     vq_model.eval()
     print(vq_model.load_state_dict(dict_s2["weight"], strict=False))
     dict_language = dict_language_v1 if version =='v1' else dict_language_v2
-    with open("./weight.json")as f:
+    with open(weight_json_file_path)as f:
         data=f.read()
         data=json.loads(data)
         data["SoVITS"][version]=sovits_path
-    with open("./weight.json","w")as f:f.write(json.dumps(data))
+    with open(weight_json_file_path,"w")as f:f.write(json.dumps(data))
     if prompt_language is not None and text_language is not None:
         if prompt_language in list(dict_language.keys()):
             prompt_text_update, prompt_language_update = {'__type__':'update'},  {'__type__':'update', 'value':prompt_language}
@@ -249,11 +259,11 @@ def change_gpt_weights(gpt_path):
     t2s_model.eval()
     total = sum([param.nelement() for param in t2s_model.parameters()])
     print("Number of parameter: %.2fM" % (total / 1e6))
-    with open("./weight.json")as f:
+    with open(weight_json_file_path)as f:
         data=f.read()
         data=json.loads(data)
         data["GPT"][version]=gpt_path
-    with open("./weight.json","w")as f:f.write(json.dumps(data))
+    with open(weight_json_file_path,"w")as f:f.write(json.dumps(data))
 
 
 change_gpt_weights(gpt_path)
@@ -303,7 +313,7 @@ def get_first(text):
     text = re.split(pattern, text)[0].strip()
     return text
 
-from text import chinese
+from GPT_SoVITS.text import chinese
 def get_phones_and_bert(text,language,version,final=False):
     if language in {"en", "all_zh", "all_ja", "all_ko", "all_yue"}:
         language = language.replace("all_","")

GPT_SoVITS/inference_webui_fast.py

@@ -46,7 +46,7 @@ version=os.environ.get("version","v2")
 import gradio as gr
 from TTS_infer_pack.TTS import TTS, TTS_Config
 from TTS_infer_pack.text_segmentation_method import get_method
-from tools.i18n.i18n import I18nAuto, scan_language_list
+from GPT_SoVITS.tools.i18n.i18n import I18nAuto, scan_language_list
 
 language=os.environ.get("language","Auto")
 language=sys.argv[-1] if sys.argv[-1] in scan_language_list() else language

GPT_SoVITS/module/attentions.py

@@ -3,8 +3,8 @@ import torch
 from torch import nn
 from torch.nn import functional as F
 
-from module import commons
-from module.modules import LayerNorm
+from GPT_SoVITS.module import commons
+from GPT_SoVITS.module.modules import LayerNorm
 
 
 class Encoder(nn.Module):

GPT_SoVITS/module/data_utils.py

@@ -17,7 +17,7 @@ from functools import lru_cache
 import requests
 from scipy.io import wavfile
 from io import BytesIO
-from tools.my_utils import load_audio
+from GPT_SoVITS.tools.my_utils import load_audio
 version = os.environ.get('version',None)
 # ZeroDivisionError fixed by Tybost (https://github.com/RVC-Boss/GPT-SoVITS/issues/79)
 class TextAudioSpeakerLoader(torch.utils.data.Dataset):

GPT_SoVITS/module/models.py

@@ -9,18 +9,18 @@ import torch
 from torch import nn
 from torch.nn import functional as F
 
-from module import commons
-from module import modules
-from module import attentions
+from GPT_SoVITS.module import commons
+from GPT_SoVITS.module import modules
+from GPT_SoVITS.module import attentions
 
 from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
 from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
-from module.commons import init_weights, get_padding
-from module.mrte_model import MRTE
-from module.quantize import ResidualVectorQuantizer
+from GPT_SoVITS.module.commons import init_weights, get_padding
+from GPT_SoVITS.module.mrte_model import MRTE
+from GPT_SoVITS.module.quantize import ResidualVectorQuantizer
 # from text import symbols
-from text import symbols as symbols_v1
-from text import symbols2 as symbols_v2
+from GPT_SoVITS.text import symbols as symbols_v1
+from GPT_SoVITS.text import symbols2 as symbols_v2
 from torch.cuda.amp import autocast
 import contextlib
 

GPT_SoVITS/module/modules.py

@@ -7,9 +7,9 @@ from torch.nn import functional as F
 from torch.nn import Conv1d
 from torch.nn.utils import weight_norm, remove_weight_norm
 
-from module import commons
-from module.commons import init_weights, get_padding
-from module.transforms import piecewise_rational_quadratic_transform
+from GPT_SoVITS.module import commons
+from GPT_SoVITS.module.commons import init_weights, get_padding
+from GPT_SoVITS.module.transforms import piecewise_rational_quadratic_transform
 import torch.distributions as D
 
 

GPT_SoVITS/module/mrte_model.py

@@ -3,7 +3,7 @@
 import torch
 from torch import nn
 from torch.nn.utils import remove_weight_norm, weight_norm
-from module.attentions import MultiHeadAttention
+from GPT_SoVITS.module.attentions import MultiHeadAttention
 
 
 class MRTE(nn.Module):

GPT_SoVITS/module/quantize.py

@@ -13,7 +13,7 @@ import typing as tp
 import torch
 from torch import nn
 
-from module.core_vq import ResidualVectorQuantization
+from GPT_SoVITS.module.core_vq import ResidualVectorQuantization
 
 
 @dataclass

GPT_SoVITS/onnx_export.py

@@ -10,7 +10,7 @@ cnhubert.cnhubert_base_path = cnhubert_base_path
 ssl_model = cnhubert.get_model()
 from text import cleaned_text_to_sequence
 import soundfile
-from tools.my_utils import load_audio
+from GPT_SoVITS.tools.my_utils import load_audio
 import os
 import json
 

GPT_SoVITS/prepare_datasets/1-get-text.py

@@ -21,7 +21,7 @@ from tqdm import tqdm
 from text.cleaner import clean_text
 from transformers import AutoModelForMaskedLM, AutoTokenizer
 import numpy as np
-from tools.my_utils import clean_path
+from GPT_SoVITS.tools.my_utils import clean_path
 
 # inp_text=sys.argv[1]
 # inp_wav_dir=sys.argv[2]

GPT_SoVITS/prepare_datasets/2-get-hubert-wav32k.py

@@ -19,7 +19,7 @@ from scipy.io import wavfile
 import librosa
 now_dir = os.getcwd()
 sys.path.append(now_dir)
-from tools.my_utils import load_audio,clean_path
+from GPT_SoVITS.tools.my_utils import load_audio,clean_path
 
 # from config import cnhubert_base_path
 # cnhubert.cnhubert_base_path=cnhubert_base_path

GPT_SoVITS/prepare_datasets/3-get-semantic.py

@@ -24,7 +24,7 @@ from glob import glob
 from tqdm import tqdm
 import logging, librosa, utils
 from module.models import SynthesizerTrn
-from tools.my_utils import clean_path
+from GPT_SoVITS.tools.my_utils import clean_path
 logging.getLogger("numba").setLevel(logging.WARNING)
 # from config import pretrained_s2G
 

GPT_SoVITS/process_ckpt.py

@@ -3,7 +3,7 @@ from collections import OrderedDict
 from time import time as ttime
 import shutil,os
 import torch
-from tools.i18n.i18n import I18nAuto
+from GPT_SoVITS.tools.i18n.i18n import I18nAuto
 
 i18n = I18nAuto()
 

GPT_SoVITS/text/__init__.py

@@ -4,8 +4,8 @@ import os
 # else:
 #   from text.symbols2 import symbols
 
-from text import symbols as symbols_v1
-from text import symbols2 as symbols_v2
+from GPT_SoVITS.text import symbols as symbols_v1
+from GPT_SoVITS.text import symbols2 as symbols_v2
 
 _symbol_to_id_v1 = {s: i for i, s in enumerate(symbols_v1.symbols)}
 _symbol_to_id_v2 = {s: i for i, s in enumerate(symbols_v2.symbols)}

GPT_SoVITS/text/chinese.py

@@ -5,9 +5,9 @@ import re
 import cn2an
 from pypinyin import lazy_pinyin, Style
 
-from text.symbols import punctuation
-from text.tone_sandhi import ToneSandhi
-from text.zh_normalization.text_normlization import TextNormalizer
+from GPT_SoVITS.text.symbols import punctuation
+from GPT_SoVITS.text.tone_sandhi import ToneSandhi
+from GPT_SoVITS.text.zh_normalization.text_normlization import TextNormalizer
 
 normalizer = lambda x: cn2an.transform(x, "an2cn")
 

GPT_SoVITS/text/chinese2.py

@@ -6,9 +6,9 @@ import cn2an
 from pypinyin import lazy_pinyin, Style
 from pypinyin.contrib.tone_convert import to_normal, to_finals_tone3, to_initials, to_finals
 
-from text.symbols import punctuation
-from text.tone_sandhi import ToneSandhi
-from text.zh_normalization.text_normlization import TextNormalizer
+from GPT_SoVITS.text.symbols import punctuation
+from GPT_SoVITS.text.tone_sandhi import ToneSandhi
+from GPT_SoVITS.text.zh_normalization.text_normlization import TextNormalizer
 
 normalizer = lambda x: cn2an.transform(x, "an2cn")
 
@@ -25,9 +25,9 @@ import jieba_fast.posseg as psg
 is_g2pw = True#True if is_g2pw_str.lower() == 'true' else False
 if is_g2pw:
     print("当前使用g2pw进行拼音推理")
-    from text.g2pw import G2PWPinyin, correct_pronunciation
+    from GPT_SoVITS.text.g2pw import G2PWPinyin, correct_pronunciation
     parent_directory = os.path.dirname(current_file_path)
-    g2pw = G2PWPinyin(model_dir="GPT_SoVITS/text/G2PWModel",model_source=os.environ.get("bert_path","GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large"),v_to_u=False, neutral_tone_with_five=True)
+    g2pw = G2PWPinyin(model_dir=f"{current_file_path}/G2PWModel",model_source=os.environ.get("bert_path","GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large"),v_to_u=False, neutral_tone_with_five=True)
 
 rep_map = {
     "：": ",",

GPT_SoVITS/text/cleaner.py

@@ -1,4 +1,4 @@
-from text import cleaned_text_to_sequence
+from GPT_SoVITS.text import cleaned_text_to_sequence
 import os
 # if os.environ.get("version","v1")=="v1":
 #     from text import chinese
@@ -7,8 +7,8 @@ import os
 #     from text import chinese2 as chinese
 #     from text.symbols2 import symbols
 
-from text import symbols as symbols_v1
-from text import symbols2 as symbols_v2
+from GPT_SoVITS.text import symbols as symbols_v1
+from GPT_SoVITS.text import symbols2 as symbols_v2
 
 special = [
     # ("%", "zh", "SP"),
@@ -33,7 +33,7 @@ def clean_text(text, language, version=None):
     for special_s, special_l, target_symbol in special:
         if special_s in text and language == special_l:
             return clean_special(text, language, special_s, target_symbol, version)
-    language_module = __import__("text."+language_module_map[language],fromlist=[language_module_map[language]])
+    language_module = __import__("GPT_SoVITS.text."+language_module_map[language],fromlist=[language_module_map[language]])
     if hasattr(language_module,"text_normalize"):
         norm_text = language_module.text_normalize(text)
     else:

GPT_SoVITS/text/g2pw/__init__.py

@@ -1 +1 @@
-from text.g2pw.g2pw import *
+from GPT_SoVITS.text.g2pw.g2pw import *

GPT_SoVITS/text/zh_normalization/__init__.py

@@ -11,4 +11,4 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-from text.zh_normalization.text_normlization import *
+from GPT_SoVITS.text.zh_normalization.text_normlization import *