-
-
Notifications
You must be signed in to change notification settings - Fork 8
/
Copy pathCHANGELOG.txt
71 lines (65 loc) · 3.39 KB
/
CHANGELOG.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
v001
- Initial release
- Supports only English SOFA Alignment with tgm_sofa
v002
- Fixed linux icon bug
- Updated tgm_sofa to v004
- Changed SOFA inference mode to force from match (gives better results to unseen speakers)
v003
- SOFA_GUI now only has full functionality on Windows
- this is because of the Japanese G2p relying on a precompiled '.exe' version of OpenJTalk.
- Fully supports Japanese SOFA model by colstone.
- Japanese transcripts are phoneme level, as the best results came from using a G2p as opposed to a dictionary.
- Installation is now 1 step, the "installer.bat" file!
- Huge code overhaul, including:
- better pipeline for changing display language (destroys the window and rebuilds it)
- put Whisper + SOFA functions into their own file to keep the code readable
- removed the need for aria2c and unzip by using gdown and zipfile
- readme file is a bit more clear
- removed the progress bar because it doesn't work how I want it to :(
v010
- Renamed from "SOFA_GUI" to "LabelMakr"!!!
- Fixes to code
- Changed some SOFA code to have it read ".txt" instead of ".lab"
- GUI overhauls
- Logo
- Custom font
- Fixed a lot of GUI related code
- Added Transcription editor
- Added "use tensorcores" in settings! This can speed up inference if your GPU has tensorcores.
- Implemented suco's Mandarin singing model!
v020
- Included a precompiled version of ffmpeg to avoid a "cannot find files" error that a lot of people were having.
- Upgraded included version of SOFA to 1.0.2
- not huge changes, but included some bug fixes.
- labels will now export to "corpus/{speaker}/labels" as opposed to "corpus/{speaker}/htk/phones"
- Added SOFA model & GUI translation for FRENCH!
- uses a Millefeuille SOFA model developed by @imsupposedto.
- GUI translation by Hibya and Mim.
- French transcriptions automatically account for contractions (j' in j'taime)
- Implemented the tenatively named "OovG2P" into SOFA.
- this utilizes OpenUTAU's G2P model system to guess the pronunciation for words that are not in the given dictionaries.
- This is only applied to English and French, since Japanese has it's own weird G2p thing and Chinese... doesn't really need one.
v021
- Fixed a few scripts to allow for early CPU inference.
v030
- Huge update
- Support for colstone's Korean SOFA model!
- Korean GUI translation by 군곰 (KUNGOM)
- Lot's of small GUI tweaks/changes/addition
- Light/Dark mode toggle (Dark mode by default)
- cute little icons on the buttons, custom fonts for English/Japanese/Chinese/Korean
- Allows you to choose to install CPU or GPU version
- More code consistency
- Global implementation of Pathlib
- removed weird inconsistencies, like unused variables
- Implemented the logging module for slightly easier troubleshooting
- Implemented a python module I wrote called "ez-localizr", to make localization easy :)
- First (beta) implementation of Labbu label fixing.
- uses label tool I wrote called "labbu" to contextually post-process labels
- Currently it's most useful for English, but there's still some good fixes for non-English languages.
- Fixed up the transcription editor
- Less unhelpful text in the file selection box
- Better interfacing with pygame mixer for playing audio
- Much easier to implement custom SOFA models.
- Here's a guide! https://github.com/spicytigermeat/LabelMakr/blob/v030/DOCS/implement_custom_sofa_model.md