Why do we want to classify a signal?

First things first, we’d like to introduce the use case and the data we’ll work with in this chapter. Using a microphone, we recorded audio data from four different audio sources. We’ll build a model to analyze short examples of these sounds and identify them automatically.

Our use case is focused on audio data, but the pipeline we will create works equally well with many other recordings from IoT devices. In a manufacturing setting, for example, you may use this approach to attach vibration sensors to a machine and classify the activity, or even predict anomalies for predictive maintenance tasks. The Fourier transform really can be used in any place where data has temporal, or even spatial, relations.

Let’s look at what the raw audio signal from source one looks like in the following plot.

Figure 8.1 – Plot of raw audio data from source one

The preceding plot shows a short...