11th April 2022

Initial Experiments

Experimenting with mic:

Did several tests to examine the limits of the algorithm when used with the mic.

Objective: We were trying to see if the algorithm was able to identify different sound frequencies, texture and quality accurately.

Test 1:

Drumming out a rhythm with fingers vs textural sound.

Result: The algorithm was not so good in identifying rhythm presumably because the samples saves were too short. It was able to identify textural sounds better presumably because each sample (1 second long) was enough to contain textural sound quality since this is not a time based quality

Test 2:

Using a hip-hop track vs an ambient track as sound sample

Result: Again, the algorithm was unable to decipher rhythm but could identify texture well

Test 3:

Using a synth to create two notes of different frequencies

Result: Algorithm could identify the two frequencies very efficiently, presumably because the inputs were so precise.

Test 4:

Using voice: male and female

Disclaimer: we understood how compartmentalizing voices as male and female is inherently bias as people with deeper voices can identify as women and vice-versa. However, we wanted to see if the algorithm based on a recorded sample (1 male, 1 female) could identify other voices into male and female with some accuracy.

Scenario 1: We used a sound of the vowel “A” recorded by a female and a male.