Everyone in this world (or mostly everyone :P) has a sense of rhythm in the music they listen to. This is the rhythm they dance to; this is the rhythm they sing out while humming the song. So our idea is to somehow enable users to be able to express this rhythm via the manner of beat boxing.
People have made a lot of fun of us (on multiple occasions) while we were completely immersed in beat boxing to our favorite song that came on the radio. This mockery was justified, since in comparison to professional beat boxers, our skills were no where near. So we decided, why not make a new game that helps novices like us learn how to beat box better! And that gave birth to BeatBox Hero!
Based on our premise and motivation, we wanted to build a software that helped users to beat box better and more than that enjoy the whole experience of beat boxing to a song (and not be mocked at!). Deriving inspiration from products such as Tap Tap Revenge and Guitar Hero, we decided to make this a game that people would love playing as well as enhance their beat boxing skills. We experimented with a few ideas and finally settled down to this one.
The product is a game. It initially allows the user to pick a song they would like to beat box to. This screen looks like
Once a song is selected the game begins. As shown below, the user sees a stream of balls coming down that symbolize beat sounds. Whenever a ball enters the end zone (the grey shaded region), the user is expected to make the sound corresponding to it. We keep a percentage count of the number of beats hit.
The system has 2 major components to it - the beat detection engine and the graphics visualization.
Beat Detection Engine
The purpose of the engine is to be able to detect which beat has been played when a person beat boxes to a song. There are 2 major problems we dealt with here
- Onset detection: The first problem is to figure out when a beat occurs given a stream of audio input. We solved this by basically checking whether the buffer crossed an energy threshold. Furthermore, since we wanted the product to be real time, we just store a small fraction of the beat and work with that so that we can in real time play the beat back.
- Beat pattern matching: Given an audio sample of a beat, we solve the problem of detecting which beat it is. We have 4 kinds of beats possible - bass, midtom, snare and hi-hat. We used audio features to be able to classify which one of the 4 a beat is. Since, the software is real time, the amount of information stored is very less and so complex features could not be used to distinguish between beat sounds. We used simple features such as Zero Crossings, Spectral Centroid and Pitch to distinguish between the beat sounds.
This product can be tested by picking a song with known beats and recording people exactly replicating the same beat pattern in their own voice. Then, we can measure the similarity between the generated beat and the original beat
Team of Two
1. Rohan Jain (firstname.lastname@example.org)
2. Ankit Gupta (email@example.com)
Nov, 9: Detect general beat pattern and tempo from the user sample
Nov 16: Map different styles of notes in the user sample to different percussion instruments
Nov 21: Figure out how the choose a "start" and an "end" for a vocal beat.
Nov 30: Make UI/Visualization
Dec 7: Polish UI and general project.