[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Intermediate representation for music analysis


In Chapter 4 of my 2005 dissertation "The Phonetics and Phonology of Glottal Manner Features" (pp. 93-96),


I used an auditorily-based filter bank with 330 filters spanning a 9-octave range 12.69-6501.99 Hz. The minimum separation interval between filters on an equally spaced log2 scale is 2**(1/48) octave or one-eighth tone. The linear filters are second-order, with their 3-dB bandwidths being adjusted to match Glasberg and Moore's 1990 Equivalent Rectangular Bandwidths (ERB = 24.7[4.37F + 1], where F is the center frequency in kHz). One-eighth tone (1.45%) is on the order of the smallest frequency difference limen for the second formant (F2) in speech (Kewley-Port and Watson, 1994).

This computationally simple analysis method, or a modification thereof, might be useful to you in music as well.


Quoting Ilya Sedelnikov <ilyas@xxxxxxxxxxxxxx>:

Dear list,

Does someone aware of works that use filterbanks with more than a couple of
dosens filters as a front-end for the music analysis ?

Human ear is able to distinguish pitch differences at least twice less than a
semitone which implies that for the analysis of musical piece that spans 4
octaves the number of filters should be of the order of couple hundreds.
Nevertheless front-ends commonly used for music analysis usually use not more
than couple of dosens of filters (Fourier bins), sometimes even
non-logarithmically spaced.

I will be glad to hear any opinions on the subject.

---------------------------------------------------------------- This message was sent using IMP, the Internet Messaging Program.