|
--- |
|
datasets: |
|
- mesolitica/TTS |
|
language: |
|
- ms |
|
--- |
|
|
|
# StyleTTS2 MS |
|
|
|
Forked at https://github.com/malaysia-ai/StyleTTS2-MS, only trained on first stage. |
|
|
|
## Pre-trained modules |
|
|
|
1. Forked original [yl4579/AuxiliaryASR](https://github.com/yl4579/AuxiliaryASR) at [malaysia-ai/AuxiliaryASR-Phonemizer](https://github.com/malaysia-ai/AuxiliaryASR-Phonemizer) to use `ms` phonemizer and trained on [mesolitica/tts-combine-annotated](https://huggingface.co/datasets/mesolitica/tts-combine-annotated) dataset. |
|
2. Forked original [PL-BERT](https://arxiv.org/abs/2301.08810) at [malaysia-ai/PL-BERT-MS](https://github.com/malaysia-ai/PL-BERT-MS) to use custom word tokenizer and pretrained on Malay Wikipedia and local news. |
|
|
|
## Checkpoints |
|
|
|
We uploaded full checkpoints with optimizer states at [checkpoints-first-stage](checkpoints-first-stage). |
|
|
|
## Dataset |
|
|
|
We train on [Mesolitica/TTS](https://huggingface.co/datasets/mesolitica/TTS). |