Week 1: openEAR

As the Summer of Code began, the first target set was to get a few results from openEAR as an emotion recognizer from news audio.

1. Obtaining openEAR:

openEAR is the Munich Open-Source Emotion and Affect Recognition toolkit created by Florian Eyben, Martin Woellmer and Bjoern Schuller at the Institute for Human-Machine Communication, Technische Universitaet Muenchen. It provides audio feature extraction implemented in C++ and classifiers and pre-trained models along with some perl scripts to make it flexible for tailor making one's own emotion recognizer.

It is hosted on SourceForge and can be found here. Since I am working on Ubuntu, I will provide instructions for a debian-based OS. Download the tar and decompress using:

mv ~/Downloads/openEAR-0.1.0.tar.gz ~/ 
tar -zxvf openEAR-0.1.0.tar.gz

2. Installing openEAR:

Before we install openEAR, let's get the dependencies (just in case they are not already present)

sudo apt-get install autotools automake build-essential libtool libpthread-stubs0-dev libc6-dev build-essential

It's quite possible most of these would be present on your machine. But ensure you've done these as a precaution.
Then change directory into the openEAR folder and run the following. Ensure you have execute permissions for the files.

./autogen.sh
./autogen.sh (For some reason, it works only when run twice)
./configure
make
sudo make install

3. Post Installation:

openEAR uses weka. So as a next step, let's install weka if you do not already have it.

sudo apt-get install weka

Edit: Something I recently found about openEAR is that the files fsel.pl and arff-functions.pl in the scripts folder have paths to weka coded in. So we need to change that to the paths specific to our weka installation.
 So,
$wekapath = "\$CLASSPATH:/home/don/eyb/inst/weka-3-5-6/weka.jar" should be changed to the path to the jar on your machine. I found the jar in /usr/share/java/weka.jar and used that.
Also,
$wekacmd = "java -Xmx1024m -classpath /home/don/eyb/inst/weka-3-5-6/weka.jar "; should be changed accordingly.

Although this seems quite straightforward, it took me a good amount of time and effort to get it set up right. There seem to be a lack of resources for the amateur (like me) to install openEAR and I hope that changes with this post and whatever more posts come up with regards to openEAR, making it easier for others down the line.

I spent bits and pieces of the week going through some news archives using EDGE. I watched a couple of Nancy Grace videos and The O'Reilly Factor to get an idea of what sort of speech I would be working with to predict emotions.