In another thread Rick Lyons said: "Windowing is generally used to reduce the spectral leakage that causes a strong (high amplitude) signal's spectral magnitude to cover over (swamp out) the spectral components of nearby weak signals. If a weak signal (in which you're interested) is 3-4 FFT frequency bins away from a strong signal's center frequency, use the Hamming window. If the weak signal (in which you're interested) is more than, say, 6 FFT frequency bins away from a strong signal's center frequency, use the Hanning window." I'm investigating speech. I am externally constrained to use a sample rate of 44 khz. My experimental procedure specifies looking at 10 mSec, .1 Sec, and 1 Sec samples. My thesis statement is that "speech may be suitably characterized by total power in bands no narrower than 1/x octaves". [ My current opinion is that .3 < x < .5 ]. Since any analysis I do will be based on a sample at least 40 bins wide, can i legitimately ignore leakage? [ Aside to voice recognition lurkers, I'm asking ONLY if my approach is valid in DSP terms. ;]
Might "spectral leakage" be irrelevant?
Started by ●April 2, 2004
Reply by ●April 4, 20042004-04-04
On Fri, 02 Apr 2004 16:53:12 -0600, Richard Owlett <rowlett@atlascomm.net> wrote:>In another thread Rick Lyons said: >"Windowing is generally used to reduce the spectral leakage >that causes a strong (high amplitude) signal's spectral magnitude >to cover over (swamp out) the spectral components of nearby weak >signals. If a weak signal (in which you're interested) is >3-4 FFT frequency bins away from a strong signal's center >frequency, use the Hamming window. > >If the weak signal (in which you're interested) is more than, >say, 6 FFT frequency bins away from a strong signal's center >frequency, use the Hanning window." > >I'm investigating speech. >I am externally constrained to use a sample rate of 44 khz. >My experimental procedure specifies looking at > 10 mSec, .1 Sec, and 1 Sec samples. > >My thesis statement is that "speech may be suitably characterized by >total power in bands no narrower than 1/x octaves". [ My current >opinion is that .3 < x < .5 ].Hi, It sounds like you're building some sort of fancy spectrum analyzer.>Since any analysis I do will be based on a sample at least 40 bins >wide, can i legitimately ignore leakage?40 bins huh. I'll stick my neck out and say that I don't think windowing will make much of a difference.>[ Aside to voice recognition lurkers, I'm asking ONLY if my approach >is valid in DSP terms. ;]The good news is, Rich, once you get your entire process working properly, you can test to see what effect windowing the time samples has. Good Luck, [-Rick-]
Reply by ●April 4, 20042004-04-04
Rick Lyons wrote:> On Fri, 02 Apr 2004 16:53:12 -0600, Richard Owlett > <rowlett@atlascomm.net> wrote: > > >>In another thread Rick Lyons said: >>"Windowing is generally used to reduce the spectral leakage >>that causes a strong (high amplitude) signal's spectral magnitude >>to cover over (swamp out) the spectral components of nearby weak >>signals. If a weak signal (in which you're interested) is >>3-4 FFT frequency bins away from a strong signal's center >>frequency, use the Hamming window. >> >>If the weak signal (in which you're interested) is more than, >>say, 6 FFT frequency bins away from a strong signal's center >>frequency, use the Hanning window." >> >>I'm investigating speech. >>I am externally constrained to use a sample rate of 44 khz. >>My experimental procedure specifies looking at >> 10 mSec, .1 Sec, and 1 Sec samples. >> >>My thesis statement is that "speech may be suitably characterized by >>total power in bands no narrower than 1/x octaves". [ My current >>opinion is that .3 < x < .5 ]. > > > Hi, > > It sounds like you're building some sort > of fancy spectrum analyzer.A spectrum analyzer was what started my train of thought. Doubt anyone will want to use my final "thing" as one though.> > >>Since any analysis I do will be based on a sample at least 40 bins >>wide, can i legitimately ignore leakage? > > > 40 bins huh. I'll stick my neck out and > say that I don't think windowing will > make much of a difference.Your email the other day pointed out how I can run a test case.> > >>[ Aside to voice recognition lurkers, I'm asking ONLY if my approach >>is valid in DSP terms. ;] > > > The good news is, Rich, once you get your > entire process working properly, > you can test to see what effect > windowing the time samples has. > > Good Luck, > [-Rick-] > >