First of all, many thanks to the author: @CjangCjengh for the code and reference implementation! This repository is some corrections to the source project, to correct some errors, and will be improved and optimized from time to time.

• Reference: CjangCjengh/tacotron2-japanese

How to use

1. Put raw Japanese texts in ./filelists
2. Put WAV files in ./wav
3. (Optional) Download NVIDIA's pretrained model
4. Open ./train.ipynb to install requirements and start training
5. Download NVIDIA's WaveGlow model or WaveGlow model based on Ayachi Nene
6. The pre-trained tacotron2 model is shown in this.
7. Open ./inference.ipynb to generate voice(Please remember to modify the corresponding path where the model is stored！！)

• Inference.ipynb

checkpoint_path = 'tacotron2_Model\\ayachi_nene_1'#"ayachi_nene_1"
model = load_model(hparams)

if torch.cuda.is_available() :
    model.load_state_dict(torch.load(checkpoint_path)['state_dict'])
    model.to(device).eval()
else :
    model.load_state_dict(torch.load(checkpoint_path, map_location='cpu')['state_dict'])
    model.to(device).eval()

• Inference.ipynb

waveglow_path = 'W_Model\\model' #waveglow model
if hparams.cuda_enabled :
    waveglow = torch.load(waveglow_path)['model']
    waveglow.cuda().eval()#.half()
else :
    waveglow = torch.load(waveglow_path, map_location='cpu')['model']
    waveglow.to(device).eval()#.half()


for k in waveglow.convinv:
    k.float()
denoiser = Denoiser(waveglow)

Cleaners

File ./hparams.py line 30

1. 'japanese_cleaners'

Before

何かあったらいつでも話して下さい。学院のことじゃなく、私事に関することでも何でも

After

nanikaacltaraitsudemohanashItekudasai.gakuiNnokotojanaku,shijinikaNsurukotodemonanidemo.

2. 'japanese_tokenization_cleaners'

Before

何かあったらいつでも話して下さい。学院のことじゃなく、私事に関することでも何でも

After

nani ka acl tara itsu demo hanashi te kudasai. gakuiN no koto ja naku, shiji nikaNsuru koto de mo naNdemo.

3. 'japanese_accent_cleaners'

Before

何かあったらいつでも話して下さい。学院のことじゃなく、私事に関することでも何でも

After

:na)nika a)cltara i)tsudemo ha(na)shIte ku(dasa)i.:ga(kuiNno ko(to)janaku,:shi)jini ka(Nsu)ru ko(to)demo na)nidemo.

4. 'japanese_phrase_cleaners'

Before

何かあったらいつでも話して下さい。学院のことじゃなく、私事に関することでも何でも

After

nanika acltara itsudemo hanashIte kudasai. gakuiNno kotojanaku, shijini kaNsuru kotodemo nanidemo.

Models

Remember to change this line in ./inference.ipynb

sequence = np.array(text_to_sequence(text, ['japanese_cleaners']))[None, :]

Sanoba Witch

Ayachi Nene

Cleaners Classes	Model
japanese_cleaners	Model 1
japanese_tokenization_cleaners	Model 2
japanese_accent_cleaners	Model 3

Inaba Meguru

Cleaners Classes	Model
japanese_tokenization_cleaners	Model 1
japanese_tokenization_cleaners	Model 2

Senren Banka

Tomotake Yoshino

Cleaners Classes	Model
japanese_tokenization_cleaners	Model 1
japanese_phrase_cleaners	Model 2

Murasame

Cleaners Classes	Model
japanese_accent_cleaners	Model 1

RIDDLE JOKER

Arihara Nanami

Cleaners Classes	Model
japanese_accent_cleaners	Model 1