ml5 is a great tool to prototype web-based machine learning applications. It is accessible as it targets virtual web platforms that is compact without any hardware requirement, which also means it is cost effective. ml5 materials are also open-sourced for everyone who are interested.
The prototype below uses the cocossd object detection machine learning for ml5 to use the changing detected position of a person to control sound effects. By moving around in front of the webcam, the positions of x and y will change the filter effects of the sound. In this example, the x and y positions are mapped to the low pass filter in p5 js.
Below if an application of the above propotype