Utilities¶

Function utilities¶

Most of the feature transformation in nnmnkwii.preprocessing module is 2d functions f: (T, D) -> (T, D'). The following utilities can be used for extending 2d functions to 3d by applying 2d function to each 2d slice.

`apply_each2d_padded`(func2d, X, lengths, …)	Apply function for each padded 2d slice.
`apply_each2d_trim`(func2d, X, args, *kwargs)	Apply function for each trimmed 2d slice.

Files¶

Part of files were taken from CMU ARCTIC dataset. Example quetsion file was taken from Merlin.

`example_label_file`([phone_level])	Get path of example HTS-style full-context lable file.
`example_audio_file`()	Get path of audio file.
`example_question_file`()	Get path of example question file.
`example_file_data_sources_for_duration_model`()	Get file data sources for duration model training.
`example_file_data_sources_for_acoustic_model`()	Get file data sources for acoustic model training

Linear algebra¶

`cholesky_inv`(L[, lower])	Compute inverse matrix of real symmetric positive definite matrix using choleskey factorization.
`cholesky_inv_banded`(L[, width])