Clarification regarding Web Audio fingerprinting jitter

[phani-vadrevu]

This is not really a bug, but we thought this might be the right forum to ask this question as this forum has many Web Audio experts. (Please do direct us to another place if you don't think this is correct.)

We have recently conducted a study on Web Audio fingerprinting. A preliminary draft of our paper is here. One of the things we noticed in this study is that whenever we include a FFT computation in our fingerprinting technique, there was a "jitter" in the fingerprint produced. That is, if you run the web audio fingerprinting code multiple times, we would sometimes get different outputs for some users. This happened across multiple platforms such as Windows/Chrome, Windows/Firefox, Android/Chrome etc (But, the effect was much more pronounced for Chrome browsers). Below is a short diagrammatic representation of how our Web Audio fingerprinting code worked (this is directly adapted from a prior research work):

Here is an example of a fingerprinting site we built that shows multiple iterations of the Web Audio FFT fingerprint. If you load it across different browsers/OS platforms, you will see that on some, there is a jitter in the fingerprint value produced.

However, this is not the case when we do not include the FFT mechanism as we do in an alternative Web Audio fingerprinting mechanism that we call as "DC" (standing for Dynamics Compressor):

More details about this are on pages 1-4 of our paper (including Table-1). Do you have any idea why this jitter could be happening with FFT calculations?

[padenot]

@phani-vadrevu Do you have a way for us (Mozilla) to attempt to reproduce your research, in an attempt to validate any work we're doing in the area? Thanks!