Fundamental frequency estimation and supervised learning
1–6 of 6 posts
Re: Fundamental frequency estimation and supervised learning
#2There is a good way to make the harmonic product spectrum more robust to the effects of noise. You generate a synthetic spectrum starting from a histogram of zero crossing intervals that has been "smeared" back into a continuous signal with Gaussian peaks by applying a kernel density estimate. The result can then be passed through the HPS to find the fundamental. The decorrelated zero crossings caused by noise are much less problematic than working with the HPS of the original signal.
Re: Fundamental frequency estimation and supervised learning
#3Is there a way to download all the instrument samples from the University of Iowa in a single archive file?
Re: Fundamental frequency estimation and supervised learning
#4Is there a way to download all the instrument samples from the University of Iowa in a single archive file?
Unfortunately they only provide .zips for each instrument in Post-2012 category. You can download them at the bottom of each page f.x. http://theremin.music.uiowa.edu/MIS-Pitches-2012/MISEbClarin... There's no compressed file option for the Pre-2012.
It might be easier to make a scraper then downloading all of the files individually.
Re: Fundamental frequency estimation and supervised learning
#5There is a good way to make the harmonic product spectrum more robust to the effects of noise. You generate a synthetic spectrum starting from a histogram of zero crossing intervals that has been "smeared" back into a continuous signal with Gaussian peaks by applying a kernel density estimate. The result can then be passed through the HPS to find the fundamental. The decorrelated zero crossings caused by noise are mu…
Sounds interesting, do you an example of the performance you might expect from such a system? I must say that out out of the estimators I tried, HPS is the most difficult to deal with. You need information about how the harmonics are related to the fundamental and how many harmonics you expect before applying the algorithm. Maybe there are some adaptive methods that exists for this job but HPS is intrinsically ad-hoc, which is not very nice.
Re: Fundamental frequency estimation and supervised learning
#6The MPM algorithm is an even better variation on autocorrelation than YIN.
It picks the first peak of the autocorrelation after zero-crossing that is greater than k*(max peak). k~=0.9
http://www.cs.otago.ac.nz/tartini/papers/A_Smarter_Way_to_Fi...