Are the large models better than medium

[clear hornet]

Probably always a trade off, even in the case of focusing on English: latency/resources vs. accuracy. We've found that Large is more accurate than Medium on English and slower.

[dull quiver]

Thanks. So there's no added inaccuracy from it mishearing English words as another word in some other language? It gets better with the additional non-English training, rather than worse?

[clear hornet]

You mean using large non-english model instead of a medium english model?

[dull quiver]

Large multilanguage with no further training on english than medium.en, for recognition of english including obscure english accents.