While v1 model supports 15 languages, this version supports 50+ languages. However, performance on the 15 languages mentioned in v1 are reported to be a bit lower.
Note that ST has additional two layers(Pooling, Linear), that cannot be saved in any predefined model in HG.