---
license: mit
language:
- ja
datasets:
- Akjava/ja004_speech_common-voice_22khz
---
## Speakers
```
0  - ja004_speech_common-voice_22khz 
```

## Trainings settings
### Stage 1 2 3 
batch 180

### Stage4
batch 150

### Stage5
batch 80
### StageExtra1
**After This Stage Valid Loss is useless**

merge valid files for final learning.
### StageExtra2
add mora-speaks from group3f-fromscratch.
some mora/phonome is not learned yet,to boostup learn.