July 2, 2026

Deep Learning (for Audio) with Python Course Notes

Deep Learning

This resource is from valeriovelardo.com

Video 6

The below code snippet creates a basic Multilayer Perceptron Class.

What is an MLP?

A multi-layer perceptron is a foundational type of Feedforward Artificial Neural Network (NN). There are several types of NN's and the feedforward NN passes data linearly in one direction to generate output from a certain set of inputs and they have no loops.

The below code is a basic implementation of the functioning of an MLP.

The NumPy library, which is the prominent library for scientific computing especially in ML, is imported.

A class containing three functions is written serving the network's construction, activation, and feedforward pass.

The output of the MLP is a list of two elements which are in the range of 0 to 1. The reason for this is the sigmoid function. A sigmoid function is any function that has a mathematical curve of "S" shape.

Sigmoidal functions enable the model to perform non-linear classification of data that couldn't be classified by a simple straight line:

Since the exponential term $e^{-x}$ is always positive, the denominator $1+e^{-x}$ is always greater than 1, ensuring that the output $S\left(x\right)$ always stays between 0 and 1.

What does the output signify here if the data is audio?

One simplistic yet practical application of the above code is: three inputs do signify three features of an audio and the two output maps if the audio is Speech or Music.

This is a classic example of Supervised Learning.