Downloads · 30 days
0
StyleSpeech-MelGAN-vocoder-16kHz
Guan-Ting/StyleSpeech-MelGAN-vocoder-16kHz
StyleSpeech-MelGAN-vocoder-16kHz is a machine learning model from Guan-Ting. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
StyleSpeech or Meta-StyleSpeech is a model for Multi-Speaker Adaptive Text-to-Speech Generation The StyleSpeech model can be trained by official implementation (https://github.com/KevinMIN95/StyleSpeech). This MelGAN…
Access
Public
Updated Dec 17, 2021
Repo size
17.1 MB
Likes
5
Public
Repo makeup
Click a slice to open those files.
.pt17.1 MB · 100%
At a glance
- Access
- Public
- Created
- Mar 2, 2022
- Updated
- Dec 17, 2021
- SHA
- 05a00764
Model card
From the Hugging Face model README
The MelGAN vocoder for StyleSpeech
About StyleSpeech
- StyleSpeech or Meta-StyleSpeech is a model for Multi-Speaker Adaptive Text-to-Speech Generation
- The StyleSpeech model can be trained by official implementation (https://github.com/KevinMIN95/StyleSpeech).
About MelGAN vocoder
- This MelGAN vocoder is used to transform the mel-spectrogram back to the waveform.
- StyleSpeech is based on 16k Hz sampling rate, and there is no available 16k Hz multi-speaker vocoder.
- Thus I train this vocoder from scratch using Libri-TTS train-100 hour dataset. The training pipeline is the same as the official MelGAN (https://github.com/descriptinc/melgan-neurips).
- The synthesized sounds are close to the official demo with good quality.
Usage
- Please follow the official MelGAN (https://github.com/descriptinc/melgan-neurips) to load pre-trained checkpoint and convert your mel-spectrogram back to the waveform.
Training Details
- GPU: RTX 2080Ti
- Training epoch: 3000
- Created
- Mar 2, 2022
- Updated
- Dec 17, 2021