Code base for summer 2017
-----------
  Purpose
-----------

The purpose of this code release is to make it painless and easy for other researchers
to do one of two things: (1) run the hashprint live song id system, or (2) compute
audio hashprints.  (1) will be of interest to researchers who want to compare their
own live song id approach with a baseline system.  (2) will be of interest to researchers
who want to explore the use of hashprints on other tasks or in other contexts.  This
code base has specifically been developed to make these tasks easy to do.

Note that this code base is NOT the original code used to generate results in the paper 
"Known-Artist Live Song Identification Using Audio Hashprints".  The original code base 
is not portable or easy to use: the core functionality is buried under layers of code for
running experiments, it uses tools that are specific to our institute, and it is highly 
optimized for runtime performance (it is a hybrid of matlab, python, and cython).

In contrast, this code release is optimized to minimize your headache.  I have stripped
away all functionality except the most basic form of the system (e.g. no downsampling or
rescoring).  This system is meant to be lightweight and easy to adapt to new setups.  It 
should have near-duplicate performance to the HP(1) system shown in figure 3 of the above 
referenced article.


-------------------
  Getting Started
-------------------

There are two steps to get started:

1. Download CQT toolbox.  We use the matlab toolbox described in the following paper:  
Schorkhuber, C., Klapuri, N. Holighaus, and M. Dofler, ``A Matlab Toolbox for Efficient
Perfect Reconstruction Time-Frequency Transforms with Log-Frequency Resolution".
It can be downloaded here: http://www.cs.tut.fi/sgn/arg/CQT/
Add the folder to your matlab path.

2. Look at template.  The file codeTemplate.m shows how to run the live song id system
or to compute hashprints.  The template should be easy to modify to run on your dataset.


TJ Tsai
ttsai@g.hmc.edu