nnsvs.gen.correct_vuv_by_phone

nnsvs.gen.correct_vuv_by_phone(vuv, binary_dict, linguistic_features)[source]

Correct V/UV by phone-related flags in a hed file

This function allows us to control V/UV explicitly by C-VUV_Voiced and C-VUV_Unvoied flags in a hed file. This is useful when you see your trained acoustic model have lots of V/UV errors. Note that manually controlling V/UV means we are ignoring the acoustic model’s prediction. It would have negative impact in some cases, but most cases it would help workaround V/UV errors.

Parameters:

  • vuv (ndarray) – V/UV flags

  • binary_dict (dict) – binary feature dictionary

  • linguistic_features (ndarray) – linguistic features

Returns:

corrected V/UV flags

Return type:

ndarray