search

ALIZE-Speaker-Recognition / LIA_RAL

A high-level toolkit for speaker recognition, build on top of ALIZE-Core.
http://alize.univ-avignon.fr
GNU Lesser General Public License v3.0
125 stars 27 forks source link

[I-vectors tutorial] Pre-compiled and self-compiled binaries gave different results? #12

Open linhdvu14 opened 6 years ago

linhdvu14 commented 6 years ago

I'm running the "LIA_SpkDet — I-vector System" tutorial and noticed using the pre-built binaries gave different (much better) results from recompiling binaries on my machine.

Any pointers what the issue was? Thank you.

My machine: Linux Ubuntu 12.04 (also tested on MacOS Sierra 10.12.6, giving the same results i.e. different from pre-built binaries)

How I'm compiling:

  1. Spro 4.1:

    • ./configure
    • make
    • copy spro.h and system.h into include/ folder
    • copy libspro.a into lib/ folder

  2. alize-core:

    • aclocal
    • automake --add-missing
    • autoconf
    • ./configure
    • make
    • make install

  3. LIA_RAL:

    • aclocal
    • automake --add-missing
    • autoconf
    • ./configure --with-spro=[absolute path to spro] --with-alize=[absolute path to alize-core]
    • make
    • make install

My data: .wav audio files; extract features with Spro.

minD3D commented 5 years ago

Thank you so much

MSAlghamdi commented 5 years ago

You copied spro.h and system.h into include/ folder. What include folder do you mean? Is it somewhere in the LIA_RAL project? the same Q about copyinglibspro.a into lib/ folder. which lib?