/
2023_d_YT_SoVitsSVC_training.txt
145 lines (100 loc) · 5.6 KB
/
2023_d_YT_SoVitsSVC_training.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
Timestamps:
00:00 - preparations
00:18 - Step 1 Gathering your audio data
01:41 - Step 2 Setting Up Your Colab Script
02:50 - Step 3 Training Your Model
17:32 - Step 4 Synthesizing Your Audio
Links:
####
PPP quick guide
https://docs.google.com/document/d/1PDkSrKKiHzzpUTKzBldZeKngvjeBUjyTtGCOv2GWwa0
PPP Main Guide
https://docs.google.com/document/d/1y1pfS0LCrwbbvxdn3ZksH25BKaf0LaO13uYppxIQnac
(So-Vits-SVC 4.0 training program)
https://colab.research.google.com/drive/1laRNiMSgSw_SxSnuti8oWIuC--RHzAGp
(So-Vits-SVC 4.0 synthesis github code (3.0 voice models will not be compatible))
https://github.com/effusiveperiscope/so-vits-svc/tree/eff-4.0
#########
(this is old v3.0 Colab that is presented in this video, I recommend using the above newer 4.0)
https://colab.research.google.com/drive/1rCUOOVG7-XQlVZuWRAj5IpGrMM8t07pE?usp=sharing#scrollTo=-hEFFTCfZf57
#########
Video Transcript:
#
Training an So-Vits-SVC voice Model.
So you are interested in training a voice for the So-Vits-SVC.
This tutorial will guide you through the process step-by-step.
Before we begin, make sure you have access to a Google account,
as we'll be using Google Colab for this project.
#
Step 1: Gathering Your Audio Data
First, you'll need to gather the audio files you want to use for your model.
With this training script you don't need a text transcriptions of the audio files;
All you need are clean, and high quality audio files of the character's voice.
Remember the old saying, garbage in - garbage out.
Ideally, you should have at least 15 minutes of audio,
but you may get away with as little as 5 minutes.
Create a folder with the name of the character you want to turn into a model,
and place all of the audio files inside.
Make sure the folder is zipped, and upload it to your Google Drive in a new folder called "dataset".
#
Step 2: Setting Up Your Colab Script
Now it's time to set up the Colab script for training your model.
Open the script and type in the name of the folder you created in the previous step in the text cells.
Run the cells from the top to bottom.
During the training process, Colab may prompt you to replace all the audio clips.
Type "a" for all and press enter to confirm.
#
Step 3: Training Your Model
Once you reach the training cell, the Colab script will display a tensorboard,
which will update with new audio clips as the model learns after ever thousands step loops.
The script is designed to save the model every thousand steps to your Google Drive,
so you can resume training from the newest model in case of a script crash or connection problem.
I will highly recommended to babysit the training process
as after every thousand loops it will save the "G" and "D" models in your Google Drive,
as you will need to remove the older models by hand to ensure that your google drive will not run out of space.
Also keep in mind that it may take several thousand steps for the model to produce decent-quality audio, so be patient.
After the model has been trained to your satisfaction, check if the latest versions of the "G" and "D" models are in your Google Drive.
If not, edit the last cell with the Colab directory to the trained model and download them to your Google Drive.
Once all of that is done, download the "G" and "D" models and config json to your models folder inside the So-Vits-SVC directory.
#
Step 4: Synthesizing Your Audio
Now that your model is trained, it's time to synthesize the audio.
Start the synthesis program by navigating to the correct directory folder and starting the Python user interface.
cd (folder directory)
conda activate so_svc
python inference_gui2.py
Choose the trained model, and the file you want to convert,
and adjust the pitch setting until the output audio sounds like the character's voice.
Congratulations, you've successfully trained an audio conversion model voice!
------
------
------
Extras:
------
------
------
But wait, there is something else I would like to share with you.
You may be interested in using your audio dataset on different training code,
however those will definitively require text transcript.
Typically, extracting a transcript can be a time-consuming task, involving typing it manually or digging through game files or scrapping subtitles.
Fortunately, there are tools available that can automate the process, such as this automatic WAV transcript Python tool developed by me.
This tool uses the VOSK transcription model to transcribe audio files automatically, saving you time and effort.
While this model does not include punctuation, question marks, full-stops, or other symbols,
it is still quite effective at generating accurate transcriptions.
This tool can be a valuable asset to anyone working with audio transcription, making it a great addition to your toolkit.
https://github.com/AmoArt/Automatic-Wavs-Transcriber-VOSK
#
Simply download the zip from github, install those two module requirements and run the code.
pip install vosk
pip install pyaudio
And run the VOSK_Transcriber_offline_v2.py file.
Just remember to adjust the 'audioFrameHz' segment inside the "VOSK_Transcriber_offline_v2.py" file to whatever Hz your audio files are using.
Music:
Gothic The Chronicles Of Myrtana Archolos - 04 - Peaceful Land
Gothic The Chronicles Of Myrtana Archolos - 15 - Vista Points
Gothic The Chronicles Of Myrtana Archolos - 16 - Misty Valley
Gothic The Chronicles Of Myrtana Archolos - 20 - I Can See the Island
Gothic The Chronicles Of Myrtana Archolos - 23 - A Tale of a Witch
Gothic The Chronicles Of Myrtana Archolos - 34 - Myrtana Dream
BackTraxx Music Libraries - 05 - Acoustic Environments - 05
BackTraxx Music Libraries - 05 - Acoustic Environments - 27