blackteay/GPT-SoVITS
1
0
Fork 0
mirror of https://github.com/RVC-Boss/GPT-SoVITS.git synced 2025-04-06 03:57:44 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'main' into main

Browse Source
This commit is contained in:
RVC-Boss 2024-02-07 16:44:03 +08:00 committed by GitHub
commit ff3b239fd9
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
6 changed files with 714 additions and 700 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1261
GPT_SoVITS/inference_webui.py
View File

File diff suppressed because it is too large Load Diff

10
docs/cn/Changelog_CN.md
View File

@ -87,8 +87,18 @@
2-引入paddlespeech的Normalizer https://github.com/RVC-Boss/GPT-SoVITS/pull/377 修复一些问题，例如：xx.xx%(带百分号类)，元/吨 会读成 元吨 而不是元每吨,下划线不再会报错 2-引入paddlespeech的Normalizer https://github.com/RVC-Boss/GPT-SoVITS/pull/377 修复一些问题，例如：xx.xx%(带百分号类)，元/吨 会读成 元吨 而不是元每吨,下划线不再会报错
### 20240207更新
1-修正语种传参混乱导致中文推理效果下降 https://github.com/RVC-Boss/GPT-SoVITS/issues/391
2-uvr5适配高版本librosa https://github.com/RVC-Boss/GPT-SoVITS/pull/403
3-修复uvr5 inf everywhere报错的问题(is_half传参未转换bool导致恒定半精度推理，16系显卡会inf) https://github.com/RVC-Boss/GPT-SoVITS/commit/14a285109a521679f8846589c22da8f656a46ad8
todolist： todolist：
1-中文多音字推理优化 1-中文多音字推理优化
2-测试集成faster whisper ASR日文英文

53
requirements.txt
View File

@ -1,26 +1,27 @@
numpy numpy
scipy scipy
tensorboard tensorboard
librosa==0.9.2 librosa==0.9.2
numba==0.56.4 numba==0.56.4
pytorch-lightning pytorch-lightning
gradio==3.38.0 gradio==3.38.0
ffmpeg-python gradio_client==0.8.1
onnxruntime ffmpeg-python
tqdm onnxruntime
funasr>=1.0.0 tqdm
cn2an funasr==1.0.0
pypinyin cn2an
pyopenjtalk pypinyin
g2p_en pyopenjtalk
torchaudio g2p_en
modelscope==1.10.0 torchaudio
sentencepiece modelscope==1.10.0
transformers sentencepiece
chardet transformers
PyYAML chardet
psutil PyYAML
jieba_fast psutil
jieba jieba_fast
LangSegment jieba
Faster_Whisper LangSegment
Faster_Whisper

60
tools/uvr5/lib/lib_v5/spec_utils.py
View File

@ -43,8 +43,8 @@ def wave_to_spectrogram(
wave_left = np.asfortranarray(wave[0]) wave_left = np.asfortranarray(wave[0])
wave_right = np.asfortranarray(wave[1]) wave_right = np.asfortranarray(wave[1])
spec_left = librosa.stft(wave_left, n_fft, hop_length=hop_length) spec_left = librosa.stft(wave_left, n_fft=n_fft, hop_length=hop_length)
spec_right = librosa.stft(wave_right, n_fft, hop_length=hop_length) spec_right = librosa.stft(wave_right, n_fft=n_fft, hop_length=hop_length)
spec = np.asfortranarray([spec_left, spec_right]) spec = np.asfortranarray([spec_left, spec_right])
@ -78,7 +78,7 @@ def wave_to_spectrogram_mt(
kwargs={"y": wave_left, "n_fft": n_fft, "hop_length": hop_length}, kwargs={"y": wave_left, "n_fft": n_fft, "hop_length": hop_length},
) )
thread.start() thread.start()
spec_right = librosa.stft(wave_right, n_fft, hop_length=hop_length) spec_right = librosa.stft(wave_right, n_fft=n_fft, hop_length=hop_length)
thread.join() thread.join()
spec = np.asfortranarray([spec_left, spec_right]) spec = np.asfortranarray([spec_left, spec_right])
@ -230,27 +230,31 @@ def cache_or_load(mix_path, inst_path, mp):
if d == len(mp.param["band"]): # high-end band if d == len(mp.param["band"]): # high-end band
X_wave[d], _ = librosa.load( X_wave[d], _ = librosa.load(
mix_path, bp["sr"], False, dtype=np.float32, res_type=bp["res_type"] mix_path,
sr = bp["sr"],
mono = False,
dtype = np.float32,
res_type = bp["res_type"]
) )
y_wave[d], _ = librosa.load( y_wave[d], _ = librosa.load(
inst_path, inst_path,
bp["sr"], sr = bp["sr"],
False, mono = False,
dtype=np.float32, dtype = np.float32,
res_type=bp["res_type"], res_type = bp["res_type"],
) )
else: # lower bands else: # lower bands
X_wave[d] = librosa.resample( X_wave[d] = librosa.resample(
X_wave[d + 1], X_wave[d + 1],
mp.param["band"][d + 1]["sr"], orig_sr = mp.param["band"][d + 1]["sr"],
bp["sr"], target_sr = bp["sr"],
res_type=bp["res_type"], res_type = bp["res_type"],
) )
y_wave[d] = librosa.resample( y_wave[d] = librosa.resample(
y_wave[d + 1], y_wave[d + 1],
mp.param["band"][d + 1]["sr"], orig_sr = mp.param["band"][d + 1]["sr"],
bp["sr"], target_sr = bp["sr"],
res_type=bp["res_type"], res_type = bp["res_type"],
) )
X_wave[d], y_wave[d] = align_wave_head_and_tail(X_wave[d], y_wave[d]) X_wave[d], y_wave[d] = align_wave_head_and_tail(X_wave[d], y_wave[d])
@ -401,9 +405,9 @@ def cmb_spectrogram_to_wave(spec_m, mp, extra_bins_h=None, extra_bins=None):
mp.param["mid_side_b2"], mp.param["mid_side_b2"],
mp.param["reverse"], mp.param["reverse"],
), ),
bp["sr"], orig_sr = bp["sr"],
sr, target_sr = sr,
res_type="sinc_fastest", res_type = "sinc_fastest",
) )
else: # mid else: # mid
spec_s = fft_hp_filter(spec_s, bp["hpf_start"], bp["hpf_stop"] - 1) spec_s = fft_hp_filter(spec_s, bp["hpf_start"], bp["hpf_stop"] - 1)
@ -418,8 +422,8 @@ def cmb_spectrogram_to_wave(spec_m, mp, extra_bins_h=None, extra_bins=None):
mp.param["reverse"], mp.param["reverse"],
), ),
) )
# wave = librosa.core.resample(wave2, bp['sr'], sr, res_type="sinc_fastest") # wave = librosa.core.resample(wave2, orig_sr=bp['sr'], target_sr=sr, res_type="sinc_fastest")
wave = librosa.core.resample(wave2, bp["sr"], sr, res_type="scipy") wave = librosa.core.resample(wave2, orig_sr=bp["sr"], target_sr=sr, res_type="scipy")
return wave.T return wave.T
@ -506,8 +510,8 @@ def ensembling(a, specs):
def stft(wave, nfft, hl): def stft(wave, nfft, hl):
wave_left = np.asfortranarray(wave[0]) wave_left = np.asfortranarray(wave[0])
wave_right = np.asfortranarray(wave[1]) wave_right = np.asfortranarray(wave[1])
spec_left = librosa.stft(wave_left, nfft, hop_length=hl) spec_left = librosa.stft(wave_left, n_fft=nfft, hop_length=hl)
spec_right = librosa.stft(wave_right, nfft, hop_length=hl) spec_right = librosa.stft(wave_right, n_fft=nfft, hop_length=hl)
spec = np.asfortranarray([spec_left, spec_right]) spec = np.asfortranarray([spec_left, spec_right])
return spec return spec
@ -569,10 +573,10 @@ if __name__ == "__main__":
if d == len(mp.param["band"]): # high-end band if d == len(mp.param["band"]): # high-end band
wave[d], _ = librosa.load( wave[d], _ = librosa.load(
args.input[i], args.input[i],
bp["sr"], sr = bp["sr"],
False, mono = False,
dtype=np.float32, dtype = np.float32,
res_type=bp["res_type"], res_type = bp["res_type"],
) )
if len(wave[d].shape) == 1: # mono to stereo if len(wave[d].shape) == 1: # mono to stereo
@ -580,9 +584,9 @@ if __name__ == "__main__":
else: # lower bands else: # lower bands
wave[d] = librosa.resample( wave[d] = librosa.resample(
wave[d + 1], wave[d + 1],
mp.param["band"][d + 1]["sr"], orig_sr = mp.param["band"][d + 1]["sr"],
bp["sr"], target_sr = bp["sr"],
res_type=bp["res_type"], res_type = bp["res_type"],
) )
spec[d] = wave_to_spectrogram( spec[d] = wave_to_spectrogram(

28
tools/uvr5/vr.py
View File

@ -61,19 +61,19 @@ class AudioPre:
_, _,
) = librosa.core.load( # 理论上librosa读取可能对某些音频有bug，应该上ffmpeg读取，但是太麻烦了弃坑 ) = librosa.core.load( # 理论上librosa读取可能对某些音频有bug，应该上ffmpeg读取，但是太麻烦了弃坑
music_file, music_file,
bp["sr"], sr = bp["sr"],
False, mono = False,
dtype=np.float32, dtype = np.float32,
res_type=bp["res_type"], res_type = bp["res_type"],
) )
if X_wave[d].ndim == 1: if X_wave[d].ndim == 1:
X_wave[d] = np.asfortranarray([X_wave[d], X_wave[d]]) X_wave[d] = np.asfortranarray([X_wave[d], X_wave[d]])
else: # lower bands else: # lower bands
X_wave[d] = librosa.core.resample( X_wave[d] = librosa.core.resample(
X_wave[d + 1], X_wave[d + 1],
self.mp.param["band"][d + 1]["sr"], orig_sr = self.mp.param["band"][d + 1]["sr"],
bp["sr"], target_sr = bp["sr"],
res_type=bp["res_type"], res_type = bp["res_type"],
) )
# Stft of wave source # Stft of wave source
X_spec_s[d] = spec_utils.wave_to_spectrogram_mt( X_spec_s[d] = spec_utils.wave_to_spectrogram_mt(
@ -245,19 +245,19 @@ class AudioPreDeEcho:
_, _,
) = librosa.core.load( # 理论上librosa读取可能对某些音频有bug，应该上ffmpeg读取，但是太麻烦了弃坑 ) = librosa.core.load( # 理论上librosa读取可能对某些音频有bug，应该上ffmpeg读取，但是太麻烦了弃坑
music_file, music_file,
bp["sr"], sr = bp["sr"],
False, mono = False,
dtype=np.float32, dtype = np.float32,
res_type=bp["res_type"], res_type = bp["res_type"],
) )
if X_wave[d].ndim == 1: if X_wave[d].ndim == 1:
X_wave[d] = np.asfortranarray([X_wave[d], X_wave[d]]) X_wave[d] = np.asfortranarray([X_wave[d], X_wave[d]])
else: # lower bands else: # lower bands
X_wave[d] = librosa.core.resample( X_wave[d] = librosa.core.resample(
X_wave[d + 1], X_wave[d + 1],
self.mp.param["band"][d + 1]["sr"], orig_sr = self.mp.param["band"][d + 1]["sr"],
bp["sr"], target_sr = bp["sr"],
res_type=bp["res_type"], res_type = bp["res_type"],
) )
# Stft of wave source # Stft of wave source
X_spec_s[d] = spec_utils.wave_to_spectrogram_mt( X_spec_s[d] = spec_utils.wave_to_spectrogram_mt(

2
tools/uvr5/webui.py
View File

@ -19,7 +19,7 @@ for name in os.listdir(weight_uvr5_root):
uvr5_names.append(name.replace(".pth", "")) uvr5_names.append(name.replace(".pth", ""))
device=sys.argv[1] device=sys.argv[1]
is_half=sys.argv[2] is_half=eval(sys.argv[2])
webui_port_uvr5=int(sys.argv[3]) webui_port_uvr5=int(sys.argv[3])
is_share=eval(sys.argv[4]) is_share=eval(sys.argv[4])