Introduce Docker and Windows CI Workflow, Pre-commit Formatting, and Language Resource Auto-Download (#2351)

* Docker Auto-Build Workflow

* Rename

* Update

* Fix Bugs

* Disable Progress Bar When workflows triggered

* Fix Wget

* Fix Bugs

* Fix Bugs

* Update Wget

* Update Workflows

* Accelerate Docker Image Building

* Fix Install.sh

* Add Skip-Check For Action Runner

* Fix Dockerfile

* .

* .

* .

* .

* Delete File in Runner

* Add Sort

* Delete More Files

* Delete More

* .

* .

* .

* Add Pre-Commit Hook
Update Docker

* Add Code Spell Check

* [pre-commit.ci] trigger

* [pre-commit.ci] trigger

* [pre-commit.ci] trigger

* Fix Bugs

* .

* Disable Progress Bar and Logs while using GitHub Actions

* .

* .

* Fix Bugs

* update conda

* fix bugs

* Fix Bugs

* fix bugs

* .

* .

* Quiet Installation

* fix bugs

* .

* fix bug

* .

* Fix pre-commit.ci and Docker

* fix bugs

* .

* Update Docker & Pre-Commit

* fix  bugs

* Update Req

* Update Req

* Update OpenCC

* update precommit

* .

* Update .pre-commit-config.yaml

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Update Docs and fix bugs

* Fix \

* Fix MacOS

* .

* test

* .

* Add Tag Alias

* .

* fix bugs

* fix bugs

* make image smaller

* update pre-commit config

* .

* .

* fix bugs

* use miniconda

* Fix Wrong Path

* .

* debug

* debug

* revert

* Fix Bugs

* Update Docs, Add Dict Auto Download in install.sh

* update docker_build

* Update Docs for Install.sh

* update docker docs about architecture

* Add Xcode-Commandline-Tool Installation

* Update Docs

1. Add Missing VC17
2. Modufied the Order of FFmpeg Installation and Requirements Installation
3. Remove Duplicate FFmpeg

* Fix Wrong Cuda Version

* Update TESTED ENV

* Add PYTHONNOUSERSITE(-s)

* Fix Wrapper

* Update install.sh For Robustness

* Ignore .git

* Preload CUDNN For Ctranslate2

* Remove Gradio Warnings

* Update Colab

* Fix OpenCC Problems

* Update Win DLL Strategy

* Fix Onnxruntime-gpu NVRTC Error

* Fix Path Problems

* Add Windows Packages Workflow

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* .

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* WIP

* Fix Path

* Fix Path

* Enable Logging

* Set 7-Zip compression level to maximum (-mx=9)

* Use Multithread in ONNX Session

* Fix Tag Bugs

* Add Time

* Add Time

* Add Time

* Compress More

* Copy DLL to Solve VC Runtime DLL Missing Issues

* Expose FFmpeg Errors, Copy Only Part of Visual C++ Runtime

* Update build_windows_packages.ps1

* Update build_windows_packages.ps1

* Update build_windows_packages.ps1

* Update build_windows_packages.ps1

* WIP

* WIP

* WIP

* Update build_windows_packages.ps1

* Update install.sh

* Update build_windows_packages.ps1

* Update docker-publish.yaml

* Update install.sh

* Update Dockerfile

* Update docker_build.sh

* Update miniconda_install.sh

* Update README.md

* Update README.md

* Update README.md

* Update README.md

* Update README.md

* Update README.md

* Update Colab-WebUI.ipynb

* Update Colab-Inference.ipynb

* Update docker-compose.yaml

* 更新 build_windows_packages.ps1

* Update install.sh

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

GPT_SoVITS/module/data_utils.py

@@ -470,6 +470,7 @@ class TextAudioSpeakerCollateV3:
         # return ssl_padded, spec_padded,mel_padded, ssl_lengths, spec_lengths, text_padded, text_lengths, wav_padded, wav_lengths,mel_lengths
         return ssl_padded, spec_padded, mel_padded, ssl_lengths, spec_lengths, text_padded, text_lengths, mel_lengths
 
+
 class TextAudioSpeakerLoaderV4(torch.utils.data.Dataset):
     """
     1) loads audio, speaker_id, text pairs
@@ -596,7 +597,7 @@ class TextAudioSpeakerLoaderV4(torch.utils.data.Dataset):
             audio_norm, self.filter_length, self.sampling_rate, self.hop_length, self.win_length, center=False
         )
         spec = torch.squeeze(spec, 0)
-        spec1 = spectrogram_torch(audio_norm, 1280,32000, 320, 1280,center=False)
+        spec1 = spectrogram_torch(audio_norm, 1280, 32000, 320, 1280, center=False)
         mel = spec_to_mel_torch(spec1, 1280, 100, 32000, 0, None)
         mel = self.norm_spec(torch.squeeze(mel, 0))
         return spec, mel
@@ -643,7 +644,7 @@ class TextAudioSpeakerCollateV4:
         mel_lengths = torch.LongTensor(len(batch))
 
         spec_padded = torch.FloatTensor(len(batch), batch[0][1].size(0), max_spec_len)
-        mel_padded = torch.FloatTensor(len(batch), batch[0][2].size(0), max_spec_len*2)
+        mel_padded = torch.FloatTensor(len(batch), batch[0][2].size(0), max_spec_len * 2)
         ssl_padded = torch.FloatTensor(len(batch), batch[0][0].size(1), max_ssl_len)
         text_padded = torch.LongTensor(len(batch), max_text_len)
         # wav_padded = torch.FloatTensor(len(batch), 1, max_wav_len)

GPT_SoVITS/module/mel_processing.py

@@ -39,24 +39,36 @@ hann_window = {}
 
 def spectrogram_torch(y, n_fft, sampling_rate, hop_size, win_size, center=False):
     if torch.min(y) < -1.2:
-        print('min value is ', torch.min(y))
+        print("min value is ", torch.min(y))
     if torch.max(y) > 1.2:
-        print('max value is ', torch.max(y))
+        print("max value is ", torch.max(y))
 
     global hann_window
-    dtype_device = str(y.dtype) + '_' + str(y.device)
+    dtype_device = str(y.dtype) + "_" + str(y.device)
     # wnsize_dtype_device = str(win_size) + '_' + dtype_device
-    key = "%s-%s-%s-%s-%s" %(dtype_device,n_fft, sampling_rate, hop_size, win_size)
+    key = "%s-%s-%s-%s-%s" % (dtype_device, n_fft, sampling_rate, hop_size, win_size)
     # if wnsize_dtype_device not in hann_window:
     if key not in hann_window:
         # hann_window[wnsize_dtype_device] = torch.hann_window(win_size).to(dtype=y.dtype, device=y.device)
         hann_window[key] = torch.hann_window(win_size).to(dtype=y.dtype, device=y.device)
 
-    y = torch.nn.functional.pad(y.unsqueeze(1), (int((n_fft-hop_size)/2), int((n_fft-hop_size)/2)), mode='reflect')
+    y = torch.nn.functional.pad(
+        y.unsqueeze(1), (int((n_fft - hop_size) / 2), int((n_fft - hop_size) / 2)), mode="reflect"
+    )
     y = y.squeeze(1)
     # spec = torch.stft(y, n_fft, hop_length=hop_size, win_length=win_size, window=hann_window[wnsize_dtype_device],
-    spec = torch.stft(y, n_fft, hop_length=hop_size, win_length=win_size, window=hann_window[key],
-                      center=center, pad_mode='reflect', normalized=False, onesided=True, return_complex=False)
+    spec = torch.stft(
+        y,
+        n_fft,
+        hop_length=hop_size,
+        win_length=win_size,
+        window=hann_window[key],
+        center=center,
+        pad_mode="reflect",
+        normalized=False,
+        onesided=True,
+        return_complex=False,
+    )
 
     spec = torch.sqrt(spec.pow(2).sum(-1) + 1e-8)
     return spec
@@ -64,9 +76,9 @@ def spectrogram_torch(y, n_fft, sampling_rate, hop_size, win_size, center=False)
 
 def spec_to_mel_torch(spec, n_fft, num_mels, sampling_rate, fmin, fmax):
     global mel_basis
-    dtype_device = str(spec.dtype) + '_' + str(spec.device)
+    dtype_device = str(spec.dtype) + "_" + str(spec.device)
     # fmax_dtype_device = str(fmax) + '_' + dtype_device
-    key = "%s-%s-%s-%s-%s-%s"%(dtype_device,n_fft, num_mels, sampling_rate, fmin, fmax)
+    key = "%s-%s-%s-%s-%s-%s" % (dtype_device, n_fft, num_mels, sampling_rate, fmin, fmax)
     # if fmax_dtype_device not in mel_basis:
     if key not in mel_basis:
         mel = librosa_mel_fn(sr=sampling_rate, n_fft=n_fft, n_mels=num_mels, fmin=fmin, fmax=fmax)
@@ -78,17 +90,25 @@ def spec_to_mel_torch(spec, n_fft, num_mels, sampling_rate, fmin, fmax):
     return spec
 
 
-
 def mel_spectrogram_torch(y, n_fft, num_mels, sampling_rate, hop_size, win_size, fmin, fmax, center=False):
     if torch.min(y) < -1.2:
-        print('min value is ', torch.min(y))
+        print("min value is ", torch.min(y))
     if torch.max(y) > 1.2:
-        print('max value is ', torch.max(y))
+        print("max value is ", torch.max(y))
 
     global mel_basis, hann_window
-    dtype_device = str(y.dtype) + '_' + str(y.device)
+    dtype_device = str(y.dtype) + "_" + str(y.device)
     # fmax_dtype_device = str(fmax) + '_' + dtype_device
-    fmax_dtype_device = "%s-%s-%s-%s-%s-%s-%s-%s"%(dtype_device,n_fft, num_mels, sampling_rate, hop_size, win_size, fmin, fmax)
+    fmax_dtype_device = "%s-%s-%s-%s-%s-%s-%s-%s" % (
+        dtype_device,
+        n_fft,
+        num_mels,
+        sampling_rate,
+        hop_size,
+        win_size,
+        fmin,
+        fmax,
+    )
     # wnsize_dtype_device = str(win_size) + '_' + dtype_device
     wnsize_dtype_device = fmax_dtype_device
     if fmax_dtype_device not in mel_basis:
@@ -97,11 +117,23 @@ def mel_spectrogram_torch(y, n_fft, num_mels, sampling_rate, hop_size, win_size,
     if wnsize_dtype_device not in hann_window:
         hann_window[wnsize_dtype_device] = torch.hann_window(win_size).to(dtype=y.dtype, device=y.device)
 
-    y = torch.nn.functional.pad(y.unsqueeze(1), (int((n_fft-hop_size)/2), int((n_fft-hop_size)/2)), mode='reflect')
+    y = torch.nn.functional.pad(
+        y.unsqueeze(1), (int((n_fft - hop_size) / 2), int((n_fft - hop_size) / 2)), mode="reflect"
+    )
     y = y.squeeze(1)
 
-    spec = torch.stft(y, n_fft, hop_length=hop_size, win_length=win_size, window=hann_window[wnsize_dtype_device],
-                      center=center, pad_mode='reflect', normalized=False, onesided=True, return_complex=False)
+    spec = torch.stft(
+        y,
+        n_fft,
+        hop_length=hop_size,
+        win_length=win_size,
+        window=hann_window[wnsize_dtype_device],
+        center=center,
+        pad_mode="reflect",
+        normalized=False,
+        onesided=True,
+        return_complex=False,
+    )
 
     spec = torch.sqrt(spec.pow(2).sum(-1) + 1e-8)
 

GPT_SoVITS/module/models.py

@@ -414,7 +414,8 @@ class Generator(torch.nn.Module):
         upsample_rates,
         upsample_initial_channel,
         upsample_kernel_sizes,
-        gin_channels=0,is_bias=False,
+        gin_channels=0,
+        is_bias=False,
     ):
         super(Generator, self).__init__()
         self.num_kernels = len(resblock_kernel_sizes)
@@ -1173,7 +1174,7 @@ class SynthesizerTrnV3(nn.Module):
                 quantized = F.interpolate(quantized, scale_factor=2, mode="nearest")  ##BCT
                 x, m_p, logs_p, y_mask = self.enc_p(quantized, y_lengths, text, text_lengths, ge)
         fea = self.bridge(x)
-        fea = F.interpolate(fea, scale_factor=(1.875 if self.version=="v3"else 2), mode="nearest")  ##BCT
+        fea = F.interpolate(fea, scale_factor=(1.875 if self.version == "v3" else 2), mode="nearest")  ##BCT
         fea, y_mask_ = self.wns1(
             fea, mel_lengths, ge
         )  ##If the 1-minute fine-tuning works fine, no need to manually adjust the learning rate.
@@ -1196,9 +1197,9 @@ class SynthesizerTrnV3(nn.Module):
             ge = self.ref_enc(refer[:, :704] * refer_mask, refer_mask)
         y_lengths = torch.LongTensor([int(codes.size(2) * 2)]).to(codes.device)
         if speed == 1:
-            sizee = int(codes.size(2) * (3.875 if self.version=="v3"else 4))
+            sizee = int(codes.size(2) * (3.875 if self.version == "v3" else 4))
         else:
-            sizee = int(codes.size(2) * (3.875 if self.version=="v3"else 4) / speed) + 1
+            sizee = int(codes.size(2) * (3.875 if self.version == "v3" else 4) / speed) + 1
         y_lengths1 = torch.LongTensor([sizee]).to(codes.device)
         text_lengths = torch.LongTensor([text.size(-1)]).to(text.device)
 
@@ -1207,7 +1208,7 @@ class SynthesizerTrnV3(nn.Module):
             quantized = F.interpolate(quantized, scale_factor=2, mode="nearest")  ##BCT
         x, m_p, logs_p, y_mask = self.enc_p(quantized, y_lengths, text, text_lengths, ge, speed)
         fea = self.bridge(x)
-        fea = F.interpolate(fea, scale_factor=(1.875 if self.version=="v3"else 2), mode="nearest")  ##BCT
+        fea = F.interpolate(fea, scale_factor=(1.875 if self.version == "v3" else 2), mode="nearest")  ##BCT
         ####more wn paramter to learn mel
         fea, y_mask_ = self.wns1(fea, y_lengths1, ge)
         return fea, ge