https://github.com/ttop32/coqui_tts_korea
Korean TTS using coqui TTS (glowtts and multiband melgan) - 한국어 TTS
https://github.com/ttop32/coqui_tts_korea
coqui coqui-ai deep-learning glow-tts half-life korea korean korean-language korean-letters korean-text-processing korean-tokenizer korean-tts multiband-melgan pytorch speech speech-synthesis text-to-speech tts vocoder voice-cloning
Last synced: 1 day ago
JSON representation
Korean TTS using coqui TTS (glowtts and multiband melgan) - 한국어 TTS
- Host: GitHub
- URL: https://github.com/ttop32/coqui_tts_korea
- Owner: ttop32
- License: mpl-2.0
- Created: 2021-12-07T13:42:17.000Z (over 3 years ago)
- Default Branch: main
- Last Pushed: 2022-01-28T10:29:31.000Z (over 3 years ago)
- Last Synced: 2025-04-05T12:11:29.365Z (25 days ago)
- Topics: coqui, coqui-ai, deep-learning, glow-tts, half-life, korea, korean, korean-language, korean-letters, korean-text-processing, korean-tokenizer, korean-tts, multiband-melgan, pytorch, speech, speech-synthesis, text-to-speech, tts, vocoder, voice-cloning
- Language: Jupyter Notebook
- Homepage:
- Size: 2.79 MB
- Stars: 57
- Watchers: 3
- Forks: 17
- Open Issues: 2
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# coqui_tts_korea
Korean TTS using coqui TTS (glowtts and multiband melgan) - 한국어 TTS
- [colab](https://colab.research.google.com/drive/1hv37sT7Pq-qKZe9Ihbbp5XZ-A9tsURli?usp=sharing)
pretrain with KSS data
finetune HalfLife scientist data
# Result
- input text
- "신은 우리의 수학 문제에는 관심이 없다. 신은 다만 경험적으로 통합할 뿐이다."
- output
https://user-images.githubusercontent.com/46513852/146552491-76df02ca-870d-4900-ab47-e956ede4cb84.mov
https://user-images.githubusercontent.com/46513852/150669352-d1c0aaf8-915b-498e-8441-32e99651fe1a.mov
# Train detail
- glowtts
- trained with kss data 190000 step
- train ipynb file : coqui_train_glowtts.ipynb
- google drive link : https://drive.google.com/drive/folders/1quLOabjkAmmw6mFbcCsMqmGxMC4bbbCW
- multiband-melgan
- trained with korea concat data (KSS, Zeroth and Pansori-TEDxKR) 150000 step
- train ipynb file : coqui_train_mbmelgan.ipynb
- google drive link : https://drive.google.com/drive/folders/1FOlcOjx47j_ALNw28rZkr62iOWqHY6tE
- halfLife finetuned glowtts
- trained with kss data 190000 step + halfLife 90000 step
- train ipynb file : halfLife_finetune_glowtts.ipynb
- google drive link : https://drive.google.com/drive/folders/1RubvJSDKZ_hNp3xj8mCocwtWG3KBmT4R?usp=sharing
- halfLife finetuned multiband-melgan
- trained with korea concat data (KSS, Zeroth and Pansori-TEDxKR) 150000 step + halfLife 20000 step
- train ipynb file : halfLife_finetune_mbmelgan.ipynb
- google drive link : https://drive.google.com/drive/folders/15eAW8jTHSIOAisiPQa03VOMOH-pACguc?usp=sharing
# Dataset
- [KSS Dataset](https://www.kaggle.com/bryanpark/korean-single-speaker-speech-dataset)
- [Zeroth Korean](https://github.com/goodatlas/zeroth)
- [Pansori-TEDxKR](https://github.com/yc9701/pansori-tedxkr-corpus)
- [half_life_dataset](https://bbs.ruliweb.com/news/board/1003/read/1962882)
# Required Environment to run
```python
!pip install TTS
!pip install jamo
!pip install torchaudio==0.9.0
!pip install gdown
!conda install -c conda-forge kaggle -y
!pip install librosa
```
# Acknowledgement and References
- [coqui tts](https://github.com/coqui-ai/TTS)
- [TensorFlowTTS](https://github.com/TensorSpeech/TensorFlowTTS)
- [glow tts](https://arxiv.org/abs/2005.11129)
- [Multi-band MelGAN](https://arxiv.org/abs/2005.05106)
- [FastSpeech 2](https://arxiv.org/abs/2006.04558)
- [speech-japanese-korean-vietnamese](http://www.hieuthi.com/blog/2018/04/22/speech-japanese-korean-vietnamese.html)
- [openslr](http://openslr.org/index.html)
- [half_life_dataset](https://bbs.ruliweb.com/news/board/1003/read/1962882)
- [KSS Dataset](https://www.kaggle.com/bryanpark/korean-single-speaker-speech-dataset)
- [Zeroth Korean](https://github.com/goodatlas/zeroth)
- [Pansori-TEDxKR](https://github.com/yc9701/pansori-tedxkr-corpus)
- [Fine-Tuning with a small dataset](https://github.com/TensorSpeech/TensorFlowTTS/issues/296)
- [Siri를 아이유 목소리로 바꾸기](https://blog.crux.cx/iu-siri-1/)
- [인공지능 deep voice를 이용한 TTS(음성합성) 구현하기 _ 손석희 앵커](http://melonicedlatte.com/machinelearning/2018/07/02/215933.html)
- [SCE-TTS: 내 목소리로 TTS 만들기](https://gist.github.com/yunho0130/a97db3296314cd7076d8436238fa113a)
- [huggingface_fastspeech2_kss](https://huggingface.co/tensorspeech/tts-fastspeech2-kss-ko)
- [huggingface_TensorFlowTTS](https://huggingface.co/spaces/akhaliq/TensorFlowTTS)
- [jamo](https://github.com/JDongian/python-jamo)
- [korean.py](https://github.com/TensorSpeech/TensorFlowTTS/blob/master/tensorflow_tts/utils/korean.py)
- [freeconvert](https://www.freeconvert.com/wav-to-mov)