The Debian deep learning team's machine learning policy would call this a "toxic candy" model: https://salsa.debian.org/deeplearning-team/ml-policy https://sal

The Debian deep learning team's machine learning policy would call this a "toxic candy" model: https://salsa.debian.org/deeplearning-team/ml-policy https://salsa.debian.org/deeplearning-team/ml-policy BTW, wouldn't you take the existing model and do additional Hokkaido Japanese speaker training on top of it, rather than retraining the model from scratch?