Compatibilities

The implementation of SIDEKIT benefits from the experience of existing tools and toolkits
in the community. The main ones are ALIZE, BOSARIS, HTK and LIBSVM
As far as possible, SIDEKIT as been made compatible with those tools by providing read and write
functions in the appropriate formats and using similar structures.

ALIZE

SIDEKIT is able to read and write in ALIZE binary format

  • a Gaussian Mixture Model

  • a label file

  • a matrix of statistics computed by using TotalVariability.exe or ComputeJFAStats.exe.

BOSARIS

A part of the BOSARIS toolkit has been translated into Python
in order to manipulate

  • enrollment lists as IdMap objects

  • trial lists as Ndx objects

  • score matrices as Scores objects

  • trial keys as Key objects

to plot Detection Error Trade-off (DET) curves
and compute minimum costs as defined by the NIST .

HTK

SIDEKIT is able to read and write in HTK format

  • a feature file (non-compressed)

  • a Gaussian Mixture Model (stored as a 3 states HMM)

LIBSVM

SIDEKIT makes use of the LIBSVM library [Chang11] and its Python wrapper.
High level interface are provided to train and test using SVMs.