DESCRIPTION
-----------

This is a simple project that aims to match a given audio utterance with pre-recorded ones using the
dynamical time wrapping (dtw) algorithm on mfcc or plp features.

These features (mfcc or plp) can be computed by using a open source framework like Sphinx, Kaldi or HTK.

INPUTS
------
2 features files are required as input.

Features files should be formated as follows:
:
feat0_t0, feat1_t0, feat2_t0, ..., featN_t0
feat0_t1, feat1_t1, feat2_t1, ..., featN_t1
...
feat0_tM, feat1_tM, feat2_tM, ..., featN_tM

with N the number components of the feature vector
and M the number of feature vectors contained in the file

OUTPUTS
------

The output is graphical : the dtw matrix and the a plot showing how the utterances can
be time wrapped to match.

DEPENDANCES
-----------

- QT and BOOST

FILES
-----

The interesting part (dtw compuation) is contained in UtteranceComparator{.h, .cpp} files.
These 2 files do not depend on QT.

The rest of the files are here to provide a simple graphical interface to visualize
the match between the utterances and depend on QT.

SAMPLE FILES
------------

The repository samples/ contains features (mfcc) files computed with Kaldi toolkit.
These files can be used as inputs.

SCREENSHOTS
-----------

The repository screenshots/ shows the expected outputs when 2 utterances matches or mismatches.