AnalyserNode: efficiency improvements to FFT post-processing about web-audio-api-v2 HOT 3 CLOSED

I do not know the history of why things are computed the way they are, but I suspect the original purpose of the AnalyserNode was primarily for visualization, not really analysis of the audio data. If I were doing this, I'd at least have smoothing done based on Bartlett's method of averaged periodograms.

I think it's too late to change the algorithms unless it gives something truly and significantly better in some way.

from web-audio-api-v2.

I totally agree with your assessment - I just wanted to write down the result of my looking into this here for future reference, in case someone does decide to overhaul AnalyserNode at some point, or create a better replacement for it. I can understand if there is no urgent motivation to do this any time soon though.

My parallel feature request (to get access to the raw FFT output) would provide anyone who wants to use AnalyserNode's FFT with an easy way to replace its prepackaged postprocessing with their own; I think that's far more important than spiffing up the bundled package, which seems to be designed for just a single use case.

Interestingly, Bartlett's method also smoothes the squared magnitude, same as I proposed. For a stream-oriented framework, I think replacing an average with an exponential running average as done here for smoothing is a reasonable modification.

from web-audio-api-v2.

Teleconf: We have no plans on changing the algorithm for the AnalyserNode; it's too late for that.

Closing.

from web-audio-api-v2.