Comparing features (features (MFCC, Mel spectrogram, chroma-stft, chroma-cens, and chroma-cq)
of UltraSound8k dataset (you must plot 4 histogram plots for each feature so you will end up with 20 plots)
classify the data to two classes (gunshot class and no gunshot class)
1/ find the gunshot centroid. Then what we need is a similar plot between the gunshot centroid and All the data instances in the gunshot class that will be your first plot. The second plot will be between gunshots, centroid and all the data instances in non-gunshot.
2/ Similarly find the non-gunshots centroid. And then find the similarity plot between the no gunshot centroid and Gunshot data and then the similarity between non gunshots Centroid and non-gunshot data. So that means you will have four different plots for each feature (MFCC, Mel spectrogram, chroma-stft, chroma-cens, and chroma-cq)
To plot 1 & 2:
use cosine similarity,
( if the value is 1 right, that means they are very similar, And if it is close to zero, that means they are not similar.) then find the distance. in distance, do the co
elation coefficient.
3/decoration the result(plots):
a) choose a best title for each plot to let who see your results understand what you did and what is this plot means. The title must give you an idea of what is happening in the plot? so it must be informative title. When you generate a plot; you should first look at the fonts size. If I put this font, I mean if I put this figure in a paper. Will that have a big enough font size? That's the first question that you should be asking and then.
) Then is my X axis, is my Y axis, legend, X label and Y label. Do they make sense?
c) your title should be something such as “co
elation between Gunshot centroid and gunshot data.” then for the next plot, it'll be “co
elation between gunshot centroid and no gunshot data”, something like that. (Create your own title like my example, you have to change the title what in the code) So, you can probably write the title in like 2 lines if your title is long..
d) In your Y axis your and X axis your values:
all the plots have to have same scale to make the figures are comparable, can compare them. So again, depending upon what entity you're finding, make sure you have the entire scale. So, the way to make all the plots are comparable is all the plots must have the same X axis, and Y axis.
XXXXXXXXXXmake all the plots in Y axis from -1until 1. And make all the X axis from -1 to 1 in all the 20 plots.
4/ then what is a best feature and explain why in text block.
5/ run the program in the google colab, then download the .ipynb file to submit this file with the result.