Re: formant filtering (Eckard Blumschein )

Subject: Re: formant filtering
From:    Eckard Blumschein  <Eckard.Blumschein(at)E-TECHNIK.UNI-MAGDEBURG.DE>
Date:    Thu, 27 Feb 2003 08:44:00 +0100

Dear Reddy Sivaprasad, experts of auditory physiology, and engineers: please don't take it amiss if I admit my guess: A narrower bandwidth at the level of auditory nerve would not improve but on the contrary degradate accuracy of hearing in general. You and perhaps nearly all other researcher are sticking yet in the imagination that high frequency resolution is the key. I don't think so. There is mounting evidence for a crucial role of accurate correlation instead. Roughly speaking, correlation equals the spectrum of the square or absolute value of the spectrum of original signal. In order to get a narrow spike in the (auto-) correlation function, the corresponding pulse within the spectrum of the original signal must be as wide as possible. You might compare the up to about 30 simultaneously possible critical bands (including formants) with a radar signal's stepped frequency structure. Notice, there is a frequency chirp within each single pulse or CB, respectively. Is this chirp already performed at the basilar membrane? No. As Tianying Ren has shown, there is a standing traveling wave in case of tonal stimulation. Consequently, within the cochlear nucleus a piece of tonotopy which corresponds to the CB is stimulated with one single frequency, although one would expect a CB-wide range of frequencies. My explanation to this paradox is simple and eye-opening: The belonging neurons, more exactly speaking, an estimated 33 000 asynchronously chopping T-multipolar cells are actually responding to place, not to the frequency of firing in the auditory nerve. At some higher level, there must be neurons which perform the second spectral analysis. Given a CB has settled at position 1000 to 1200 Hz, then all pertaining neuron in the auditory nerve might phase lock at 1000 Hz. The multipolars of ventral CN will perhaps chopper over a certain range of lower frequency, e.g. 100 to 105 Hz or even randomly distributed. This way, place information is tempotally processed and finally mapped according to correlation, not to the spectrum. How to improve accuracy? A natural solution is the acoustic fovea, that is, a non-linearity of frequency-place relationship in the cochleae of some bats and mole rats. A small frequency subrange is spread over a wider distance. So the second spectral analysis is based on much more approached neurons in the CN then else. The purposes of this trick are presumably extreme acuity in echo-location or improved hearing below 1 kHz, respectively. Acoustic communication seems to be less important for these animals. In brief, the CB has its width for a good reason. It is perhaps just as wide as possible without causing damage to the option of grouping multiples of a frequency. Recently, there was an unfounded suspicion that MPEG data reduction could possibly irritate the brain. As far as I can judge, there is no reason for that because nature "deliberately" made the first spectral analysis a seemingly coarse one within the inner ear. Yours sincerely, Eckard At 21:08 26.02.2003 +0530, REDDY SIVAPRASAD wrote: > Dear list, reduce the formant band width of some consonants. I would >request for some sincere suggestions to carry out this task....any >software? Thanking you > M. Reddy Sivaprasad Junior Research Fellow, Dept. of Audiology, AIISH, >MYSORE-6 Karnataka India Ph: +91-821-514449 Fax: +91-821-510515 >STOP MORE SPAM with the new MSN 8 and get 2 months FREE*

This message came from the mail archive
maintained by:
DAn Ellis <>
Electrical Engineering Dept., Columbia University