Use the HTK toolkit to train a model to recognize the spoken numbers.

Installation of HTK toolkit differs a lot on every platform. The project comes with a Vagrantfile in order to ease the project's setup. Vagrant will provision a virtual machine containing Ubuntu 32bit and also install HTK. Refers to Vagrant's documentation for more details.

Because also of the distribution policy of the HTK, you will have to register on the website and download the package. Add the compressed package (HTK-3.4.1.tar.gz) in the provision folder.

You are now ready to run Vagrant.

vagrant up
vagrant ssh
cd /vagrant

You can create a model and train it using the command ./start-htk.sh. The final model will be stored in the folder Models/hmm15 The command will also output accuracy of the trainned model against the dev dataset.

You can specify the output file for logging the script by passing the name of the file: ./start-htk.sh cmd.log. If not file is passed the command outputs will be redirected into /dev/null.

Read the report! ;)