A usual first step in machine learning is to parameterise the signal (also called "feature extraction") and here we'll make a first attempt at that.
Frame-based analysis
The properties of speech are constantly changing over time, so we need to analyse it in short sections, called frames.
Feature vectors
We will make a first attempt at parameterising each frame, but we'll need to revisit this after learning more about the probabilistic model that will be used.
Fourier analysis
A key step in parameterising speech is to move from the time domain to the frequency domain, using Fourier analysis.
