hello friends, this is about PERCEPTUAL WEIGHTING in g729. i know that perceptual weigting is used for manipulate the noise floor with respect to the signal. can anyobody tell me how exactly it will manipulate the noise level. kindly solve my problem your sincearly -tilak email: __________________________________________________ |
|
perceptual weigting
Started by ●May 31, 2000
Reply by ●June 13, 20002000-06-13
Hi, This is my understanding. If you see the spectrum of voiced speech , it has 3-5 peaks at formant frequencies and correspoding valleys. Now, the quantization/coding noise (error) is almost white. If the noise level exceeds the signal level (which is more likely at the valleys of speech spectrum), then noise will be perceived by the human ear. To avoid this, we use the perceptual wgt filter, which shapes the noise spectrum to follow the speech spectrum. Bascially the filter shapes the noise so as to put more noise power under the peaks( of speech spectrum) and less noise at the valleys. This is making use of MASKING EFFECT of human ear. A signal with high power MASKS signals with lower power and human ear can only perceive the signal with higher power. In this case, we are making sure that the singal power is more than that of noise even at the valley frequencies, which may not be the case without the wgt. filter. If you see, even the equation of wgt. filter is derived from the LPC filter, which represents the speech spectrum. Anyone, Please correct me, if wrong or you even can add more to this. Regards Suresh --- In , tilak shetty <shettytilak@y...> wrote: > hello friends, > this is about PERCEPTUAL WEIGHTING in > g729. > i know that perceptual weigting is used for > manipulate the noise floor with respect to the > signal. > can anyobody tell me how exactly it will manipulate > the noise level. > kindly solve my problem > your sincearly > -tilak > email:shettytilak@y... > > __________________________________________________ > |
|
Reply by ●June 14, 20002000-06-14
Hi, Your understanding of Perceptual weigting is correct. I would like to add that the phenomenon of masking is widely used in audio coders like MPEG, AT&T PAC etc. to shape quantization noise in such a way that it is inaudible. Regards, Satheesh.S -----Original Message----- From: [mailto:] Sent: Wednesday, June 14, 2000 12:58 AM To: Subject: [speechcoding] Re: perceptual weigting Hi, This is my understanding. If you see the spectrum of voiced speech , it has 3-5 peaks at formant frequencies and correspoding valleys. Now, the quantization/coding noise (error) is almost white. If the noise level exceeds the signal level (which is more likely at the valleys of speech spectrum), then noise will be perceiveid by the human ear. To avoid this, we use the perceptual wgt filter, which shapes the noise spectrum to follow the speech spectrum. Bascially the filter shapes the noise so as to put more noise power under the peaks( of speech spectrum) and less noise at the valleys. This is making use of MASKING EFFECT of human ear. A signal with high power MASKS signals with lower power and human ear can only perceive the signal with higher power. In this case, we are making sure that the singal power is more than that of noise even at the valley frequencies, which may not be the case without the wgt. filter. If you see, even the equation of wgt. filter is derived from the LPC filter, which represents the speech spectrum. Anyone, Please correct me, if wrong or you even can add more to this. Regards Suresh --- In , tilak shetty <shettytilak@y...> wrote: > hello friends, > this is about PERCEPTUAL WEIGHTING in > g729. > i know that perceptual weigting is used for > manipulate the noise floor with respect to the > signal. > can anyobody tell me how exactly it will manipulate > the noise level. > kindly solve my problem > your sincearly > -tilak > email:shettytilak@y... > > __________________________________________________ > Was the salesman clueless? Productopia has the answers. http://click.egroups.com/1/4633/0/_/8885/_/960925558/ To Join: To Post: To Leave: Archives: http://www.egroups.com/group/speechcoding Other DSP-Related Groups: http://www.dsprelated.com |
Reply by ●June 15, 20002000-06-15
A good experiment is to listen to/plot the error after speech coding. It isn't white when perceptual weighting isn't used (in LPC coders in the approx < 8 kb/s range) and therefore there is little shaping going on. However, using a perceptual weighting filter limits the length of the impulse response of the weighted synthesis filter (perceptual filter followed by 1/A(z)) and the importance of this may be fairly high. In any case, the important assumption is that the coder is operating in the high rate region. If it isn't, then noise shaping doesn't really do the right thing. -- wilf satheesh wrote: > > Hi, > Your understanding of Perceptual weigting is correct. I would like to add > that the phenomenon of masking is widely used in audio coders like MPEG, > AT&T PAC etc. to shape quantization noise in such a way that it is > inaudible. > Regards, > Satheesh.S > > -----Original Message----- > From: [mailto:] > Sent: Wednesday, June 14, 2000 12:58 AM > To: > Subject: [speechcoding] Re: perceptual weigting > > Hi, > This is my understanding. > > If you see the spectrum of voiced speech , it has 3-5 peaks > at formant frequencies and correspoding valleys. Now, > the quantization/coding noise (error) is almost white. If > the noise level exceeds the signal level (which is more likely > at the valleys of speech spectrum), then noise will be perceiveid > by the human ear. To avoid this, we use the perceptual wgt filter, > which shapes the noise spectrum to follow the speech spectrum. > Bascially the filter shapes the noise so as to put more noise power > under the peaks( of speech spectrum) and less noise at the valleys. > This is making use of MASKING EFFECT of human ear. A signal > with high power MASKS signals with lower power and human ear can > only perceive the signal with higher power. In this case, we > are making sure that the singal power is more than that of noise even > at the valley frequencies, which may not be the case without the > wgt. filter. If you see, even the equation of wgt. filter is > derived from the LPC filter, which represents the speech spectrum. > > Anyone, Please correct me, if wrong or you even can add more to this. > > Regards > Suresh > > --- In , tilak shetty <shettytilak@y...> > wrote: > > hello friends, > > this is about PERCEPTUAL WEIGHTING in > > g729. > > i know that perceptual weigting is used for > > manipulate the noise floor with respect to the > > signal. > > can anyobody tell me how exactly it will manipulate > > the noise level. > > kindly solve my problem > > your sincearly > > -tilak > > email:shettytilak@y... > > |