kan-bayashi
Follow
kan-bayashi
Follow
Achievements
Achievements
Highlights
- Pro
- 1 discussion answered
Popular repositories
5,216 contributions in the last year
Less
More
Activity overview
Contributed to
espnet/espnet,
kan-bayashi/ParallelWaveGAN,
espnet/espnet_model_zoo
and 5 other
repositories
Contribution activity
August 2021
Created 466 commits in 3 repositories
Created a pull request in espnet/espnet that received 8 comments
[Not for merge] Add VITS for ESPnet2-TTS
This PR adds VITS. TODO model implementation wavenet hifigan discriminator hifigan generator posterior encoder text encoder stochastic dur…
+6,210
−141
•
8
comments
Opened 47 other pull requests in 2 repositories
espnet/espnet
30
merged
1
closed
- Fix typo in hifigan
-
Revert wrongly removed lines in
tts.sh - Update README.md
- Support joint training of text2mel and vocoder
- Support language id in VITS
- Update the TTS inference
- Refactor TTS modules in ESPnet2
- Fix k2 version in CI using conda
- Support SID and LID inputs for TTS models
- Support language ID input for espnet2 TTS
- Add cython code for VITS
- Support joint training of text2mel and vocoder
- Add duration and focus_rate in output dict
- Add missing symlink to trim_silence.py for ESPnet2
- Add CSS10 recipe for ESPnet2-TTS
- Support new g2p functions for various languages
- Add missing configs for LibriTTS recipe
- Follow up of the support speaker id input
- Support speaker id input for TTS models
- Support cleaner option in phn converter util
- Add GAN-TTS task with VITS
- Add VITS-related modules
- Add vocoder modules related to VITS
- Support linear spectrogram
- Change Text2Speech class output format
- Some pull requests not shown.
kan-bayashi/ParallelWaveGAN
16
merged
- fix typo in hifigan
- Change warn message in loading stats
- Add configs for JSUT and LibriTTS
- Fix HiFiGAN generator padding when upsample scale is odd number
- Minor update
- Add useful functions
- Update version and upload pretrained model
- Prevent pop noise at the end of audio in StyleMelGAN inference
- Support hinge loss in adversarial loss
- Update default config of HiFiGAN and StyleMelGAN
- Refactor HiFi-GAN
- Fix residual calculation in HiFiGAN generator
- Update HiFi-GAN
- Support StyleMelGAN
- Support HiFi-GAN
- Refactoring with black
Reviewed 17 pull requests in 1 repository
espnet/espnet
17 pull requests
- Support joint training of text2mel and vocoder
- Support SID and LID inputs for TTS models
- Add new trainer for GAN-based training
- Add return_complex=False for stft
- Add GAN-TTS task with VITS
- Add VITS-related modules
- Add CSS10 recipe for ESPnet2-TTS
- TTS recipes for thchs30 and aishell3
- Add JtubeSpeech Recipe
- TTS recipes for thchs30 and aishell3
- TTS recipe for J-KAC corpus
- Add SIWIS recipe
- Hubert-in-Espnet
- Add vocoder modules related to VITS
- IndicSpeech TTS Scripts
- indic tts without pretrained model
- Add JMD recipe
Created an issue in kan-bayashi/ParallelWaveGAN that received 6 comments
HiFi-GAN tuning
2021/08/05
v1
Follow official setting
Mel range in mel loss is different (full vs. 80-7600)
Log base in mel loss is different (ln vs. log10)
v2
…
6
comments
Opened 1 other issue in 1 repository
kan-bayashi/ParallelWaveGAN
1
open
416
contributions
in private repositories
Aug 2 – Aug 18