Can vad_filter be applied to the detect-language endpoint?

[yroyathon]

I'm noticing some examples where there is little to no spoken audio in the first 30 seconds of video, giving the model little to no chance in identifying the language correctly. But if the model could skip ahead to where the actual words are spoken, I'm sure it'll do a lot better in detecting which language. This might slow the performance a little, but the increased accuracy would make it worth it.