open-speech / Kaldi Io

KALDI-IO lib

Kaldi c++ IO lib (static and dynamic).

It can be used as "kaldi I/O C++ API". Using this library, ".ark,.scp" formats can be read and write, and kaldi::[Matrix|Vector|..] is available.

Compile

• requirements:
  • cmake >= 3.0
  • one of blas math lib:
    • mkl (recommended)
      • install conda, and use it to install mkl: conda install mkl (mkl is installed with conda by default)
      • when cmake, conda should be in your path
    • atlas
      • ubuntu: sudo apt-get install libatlas3-base
      • when cmake, it maynot find your atlas automatically, thus you need set the math lib path as in "#About" below.
    • Accelerate framework (need do nothing for "macOS/Darwin")
    • ...

cd kaldi-io
mkdir build && cd build
cmake -DCMAKE_INSTALL_PREFIX=.. .. # set install prefix as '../kaldi-io'
make
make install 

Results:

• kaldi-io/lib:
  • libkaldi_io_static.a
  • libkaldi_io_shared.so
• kaldi-io/include:
  • kaldi-io.h (the one big header, can be modified by need)
  • sub header dirs...

Usage

• Include the header dir and link the lib

About

• The math dependency is solved automatically from system path, by cmake find_package(cmake/Modules/FindBLAS.cmake)

  • MKL/ATLAS/Accelerate(osx) is prefered
  • if you want to set specific math lib:

  # make sure build dir is clean
  cmake -DBLAS_VENDORS=[ATLAS|MKL|OPEN|..] ..
  # additioinally, custom math lib search path can be set, for example:
  cmake -DBLAS_VENDORS=ATLAS -DBLAS_ATLAS_LIB_DIRS=.../atlas/build/lib ..
  cmake -DBLAS_VENDORS=MKL -DBLAS_MKL_LIB_DIRS=/opt/intel/mkl/lib/intel64 ..