To build a speech recognition system, you need to have skills in many disciplines in addition to software development. Currently there are several approaches, usually larger companies such as Google or Apple apply Neural Networks. During this month we were experiencing a lighter approach that is applied in more limited domains. It consists of DTW feature extraction and analysis.
We upload to Github a repository with the work done, the system extracts 12 MFCC coefficients by default for the bandwidth and K-NN algorithms are used to compare the dataset.
You can download the project in the following link.