Electrical computers: arithmetic processing and calculating – Electrical digital calculating computer – Particular function performed
Reexamination Certificate
1998-03-06
2002-05-14
Mei, Xu (Department: 2644)
Electrical computers: arithmetic processing and calculating
Electrical digital calculating computer
Particular function performed
C381S071110, C379S406080
Reexamination Certificate
active
06389440
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to an adaptive filter apparatus for correcting acoustic feedback. It has particular utility in relation to echo cancellation in a video conferencing facility.
2. Description of the Related Art
A common situation where acoustic feedback causes a problem involves a public address apparatus where speech input to a microphone is amplified before being output from a nearby loudspeaker. Since a closed loop can be formed if an acoustic path exists between the loudspeaker and the microphone, there is a possibility that the signal from the loudspeaker can be fed back to the microphone to once again be amplified and output from the loudspeaker. If the microphone is placed too close to the loudspeaker, or the output of the loudspeaker is too loud, then the gain around the closed loop can exceed unity whereupon the loudspeaker will produce the howling sound associated with acoustic feedback.
Another situation in which acoustic feedback can present a problem involves a two-way link between two sites. In this case, a combination of a loudspeaker and a microphone is provided at each site. The problem which then arises is that any feedback from the loudspeaker to the microphone at one site will cause the person speaking into the microphone at the other site to hear an echo of his or her voice. This problem is particularly acute in relation to video conferencing facilities where a long delay in the arrival of the echo is caused by the speech signal being delayed to match the time taken in compressing the video signal which accompanies the voice signal. Because this delay is considerably longer than the length of a syllable, even a relatively quiet echo will make speaking into the microphone difficult. One technique which is used to reduce acoustic feedback is to connect a filter between the loudspeaker input and the microphone output. The filter is arranged to model the acoustic path between the loudspeaker and the microphone, thereby providing a signal which is similar to that which is received at the microphone from the loudspeaker. Once the signal has been provided, it can be subtracted from the microphone output signal to provide a corrected microphone output signal without the feedback component.
A further problem associated with video conference facilities is that the acoustic path between the loudspeaker and the microphone at each site is not constant. For example, the present inventors have shown that if a person present at a video conference session leans back in his or her chair, this can have a significant effect on the nature of the acoustic path between the loudspeaker and the microphone. It is therefore known to provide an adaptive filter which attempts to adapt its model of the acoustic path between the loudspeaker and the microphone as that acoustic path changes. Normally, such adaptive filters are implemented using a finite impulse response filter whose weights are updated on the basis of the previously mentioned corrected signal.
The speed at which the filter is able to adapt its model of the acoustic path is determined, at least in part, by the amount by which the filter weights may be changed in a single update. However, if the amount by which the weights may be changed is increased in order to speed up the adaptation of the filter, the filter is then less able to model the acoustic path accurately. This is because, even if the acoustic path remains constant for a substantial length of time, the filter weights do not settle to ideal values but rather fluctuate within a wide spread of values around the ideal values. Thus, conventionally, a compromise must be made between the speed of adaptation and the fluctuation which occurs once the filter has adapted to the changed acoustic path. It has been found that the presence of the fluctuation leads to so-called “misadjustment noise”. If the loudspeaker input signal is speech then “misadjustment noise” leads to an unnatural sounding residual echo.
In addition, the speed of adaptation of conventional apparatuses has been found to depend on the nature of the loudspeaker input signal. If the input signal is speech or music, then the speed of adaptation is slower than when the input signal is white noise. Since an echo canceller in an audio-conferencing environment usually operates on a speech signal the speed of adaptation of such echo cancellers is relatively slow.
Another problem arises because the gain of the acoustic path between the loudspeaker and the microphone is usually frequency dependent. For example, it is possible that the gain around the closed loop will be −40 dB for some frequencies in the audible spectrum whilst only being −10 dB for other frequencies. In such a case, as the loudspeaker volume were increased acoustic feedback would initially occur only at those other frequencies, thereby causing a ringing noise to be heard, which would become progressively louder until the limits of the loudspeaker performance were reached. Generally, feedback will occur first at frequencies at which the closed loop gain is relatively high (high loop-gain frequencies) and then later at frequencies at which the closed loop gain is relatively low (low loop-gain frequencies). The amount by which the closed loop gain can be increased before it reaches unity (and hence ringing or howling occurs) is known as the “gain margin”. Plainly, it is desirable to increase the gain margin as far as possible, but to do this it is necessary to reduce feedback at all high loop-gain frequencies in the audible spectrum. Hitherto, many feedback cancelling apparatuses have made little improvement to the gain margin.
SUMMARY OF THE INVENTION
The present invention aims to alleviate some or all of the above problems.
According to a first aspect of the present invention there is provided a filter apparatus for correcting feedback from a loudspeaker to a microphone, said apparatus comprising:
an adaptive digital filter which, in use, receives an incoming signal and provides a modelled feedback signal for subtraction from an outgoing signal to provide a corrected outgoing signal; and
means for changing the weights of said filter in accordance with an algorithm for reducing the difference between an actual feedback signal and said modelled feedback signal, each weight change including a variable scaling factor which varies in accordance with the ratio of a first value indicative of the long-term average power of the sound being fed back to a second value indicative of the short-term average power of the sound being fed back.
In order to provide an apparatus which is sufficiently fast, many feedback cancelling apparatuses implement simplified algorithms which involve assumptions which are largely invalid for signals, such as speech or music, which have statistical properties (e.g. global variance, variance as a function of frequency) which vary significantly in the short-term. The effect of the variable scaling factor of the present invention is to control the changes to the filter weights at times when such assumptions are failing. In this way, the present invention gives the advantage that the speed of adaptation can be improved when the signal is speech or music without a concomitant increase in the misadjustment noise that occurs once adaptation has been achieved.
In preferred embodiments of the present invention, the long-term average is taken over a period longer than 250 ms—this is effective to span variations in speech power which occur between one phoneme and the next.
Preferably, the short-term average is taken over a period shorter than 25 ms.
The first and/or second value may be derived using either the incoming signal or the outgoing signal. In advantageous embodiments, the outgoing signal is used as this accounts for the nature of the acoustic path between the loudspeaker and the microphone.
Preferably, said first value represents the mean of the unsigned value of m samples of said outgoing signal previous to said weight change and said second value repre
Ensor Jonathan E
Lewis Alwyn V
British Telecommunications public limited company
Mei Xu
Nixon & Vanderhye P.C.
LandOfFree
Acoustic feedback correction does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Acoustic feedback correction, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Acoustic feedback correction will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2870813