GB2126851A - Adaptive filter - Google Patents
Adaptive filter Download PDFInfo
- Publication number
- GB2126851A GB2126851A GB08318836A GB8318836A GB2126851A GB 2126851 A GB2126851 A GB 2126851A GB 08318836 A GB08318836 A GB 08318836A GB 8318836 A GB8318836 A GB 8318836A GB 2126851 A GB2126851 A GB 2126851A
- Authority
- GB
- United Kingdom
- Prior art keywords
- noise
- band
- input signal
- signal
- adaptive filter
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B1/00—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission
- H04B1/06—Receivers
- H04B1/10—Means associated with receiver for limiting or suppressing noise or interference
- H04B1/12—Neutralising, balancing, or compensation arrangements
- H04B1/123—Neutralising, balancing, or compensation arrangements using adaptive balancing or compensation means
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03H—IMPEDANCE NETWORKS, e.g. RESONANT CIRCUITS; RESONATORS
- H03H21/00—Adaptive networks
- H03H21/0012—Digital adaptive filters
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04R—LOUDSPEAKERS, MICROPHONES, GRAMOPHONE PICK-UPS OR LIKE ACOUSTIC ELECTROMECHANICAL TRANSDUCERS; DEAF-AID SETS; PUBLIC ADDRESS SYSTEMS
- H04R2225/00—Details of deaf aids covered by H04R25/00, not provided for in any of its subgroups
- H04R2225/43—Signal processing in hearing aids to enhance the speech intelligibility
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04R—LOUDSPEAKERS, MICROPHONES, GRAMOPHONE PICK-UPS OR LIKE ACOUSTIC ELECTROMECHANICAL TRANSDUCERS; DEAF-AID SETS; PUBLIC ADDRESS SYSTEMS
- H04R25/00—Deaf-aid sets, i.e. electro-acoustic or electro-mechanical hearing aids; Electric tinnitus maskers providing an auditory perception
- H04R25/50—Customised settings for obtaining desired overall acoustical characteristics
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Filters That Use Time-Delay Elements (AREA)
Abstract
An adaptive filter 100 for adaptively filtering near stationary, relatively long duration noise in an input signal containing information such as speech whose parameters change more rapidly than the noise parameters comprises a noise analysis circuit 102 which detects the presence of noise in the input signal and identifies the parameters representing the characteristics of at least the noise independently of whether information is present in the input signal. A noise reduction circuit 104 having gain elements controlled by the identified parameters minimises noise in the output of the circuit. <IMAGE>
Description
SPECIFICATION
Adaptive filter
The invention relates to an adaptive filter for a adaptively filtering near-stationary, relatively long duration noise in an input signal containing information such as speech or music.
For the purpose of the present application, the term "near-stationary, relatively long duration noise", is hereinafter referred to as noise of the type described, and is defined as noise whose frequency spectrum does not vary with time or varies only within a narrow range over a predetermined period of time particularly with respect to the spectral parameters of the information. Vehicular and machinery noises are examples of noise of the type described. The minimum duration of noise of the type described is thus considerably longer than periods over which the spectrum of normal speech remains nearly fixed, such periods being of the order of a second. In addition, the invention is described below as being applied to filtering noise of the type described from speech, but the invention is also applicable to other information bearing signals.
US Patent No. 4,025,721 discloses a method of and means for adaptively filtering noise of the type described from speech by recognising the existence of such noise, identifying its parameters, adjusting the parameters of a filter so as to filter such noise, and applying an input signal containing such noise and speech to the filter. Termination of the noise of the type described is also recognised in order to cause the input signal to then bypass the filter.
Basic to the disclosure of the above identified patent is the necessity for regognition, an an input signal, of pauses between speech intervals. When a pause containing noise of the type described is recognised, the parameters of the noise within the pause are identified and used to adjust the parameters of the filter. Subsequently, the input signal is applied to the filter which is effective to filter the noise from subsequent speech intervals.
In accordance with the present invention, an adaptive filter for adaptively filtering nearstationary, relatively long duration noise, as hereinbefore defined, in a band-limited signal containing information such as speech whose parameters change more slowly than the noise parameters comprises a noise analysis circuit for detecting the presence of noise of the type described in the input signal and for identifying the parameters representing the characteristics of at least the noise independently of whether information is present in the input signal; and a noise reduction circuit having gain element controlled by the identified parameters for minimising noise in the output of the circuit.
In contrast with the prior art which requires the presence of information (e.g. speech) in order to sense pauses or breaks in the information neither recognition of pauses between speech intervals, nor recognition of noise termination, is required; and the hardware is extremely simple and fast.
Briefly, identification is performed in the frequency domain alone allowing a spectral filter to be used to achieve the desired filtering of the noise of the type described.
In one arrangement, an input signal containing noise of the type described and information, such as speech or music, is applied in parallel to a noise-analysis circuit that includes a plurality of bandpass filters covering the range of frequencies associated with the information. The absolute value, or a function thereof (e.g. the RMS value) of the signal in each band is obtained and smoothed, preferably by computing a running average of the absolute value in the band. In one technique, noise is assumed to be present in the input signal if a threshold is exceeded by all or by a predetermined percentage of successive minima of the smoothed signal in a band.Alternatively, the smoothed signal in each band can be sampled to enable the inverse signal-to-noise ratio (SNR-1) to be computed, the presence of noise being assumed if an SNR-1 in a band exceeds a threshold. The SNR-' in a sampling interval is computed by taking the ratio of the latest stored lowest minimum of the smoothed signal in a band to the difference between the sampled value of the smoothed signal and such minimum.
If the noise-analysis circuit detects the presence of noise in accordance with the above tests, spectral parameters of the noise, or of the noise and information are then identified by the circuit and used to control the operation of a noise-reduction circuit to which the input signal is also applied. The last mentioned circuit comprises a plurality of bandpass filters having a one-to-one correspondence with the bandpass filters of the noise-analysis circuit. Each of the bandpass filters of the noise-reduction circuit is responsive to a parameter identified from the smoothed signal in the corresponding band of the noise-analysis circuit for minimizing the effect of noise in the input signal on the output of the noise-reduction circuit minimization occuring under some continuous minimization criterion.
The noise-analysis circuit can identify either noise parameters, or noise and information parameters in a band by utilizing the smoothed signal in the band. A parameter of the noise alone in a band can be in the form of a predetermined function of the lowest one of a predetermined number of successive minima of the smoothed signal in the band. Alternatively, the noise parameter can be in the form of a predetermined function of a more recent minimum of a succession of minima if such recent minimum does not exceed the previous lower minimum by more than a given percentage (e.g., 25%).
On the other hand, a parameter of the noise and information in a band can be in the form of a predetermined function of the signal-to-noise ratio (SNR) in the band. The SNR in a band is the difference between samples of the smoothed signal in the band and the most recently accepted noise-related minimum of a time-sequence of minima of the smoothed signal. The functions referred to above may be continuous functions of the variables to yield either spectral parameters or their time domain equivalents.
In the preferred form of the invention, the bandpass filters of the noise-reduction circuit have gain elements controlled by the parameters identified by the noise-analysis circuit. When noise parameters alone are identified, the gain-toparameter relationship necessary to minimize noise may be similar to or approximate a Wiener filter with a-priori knowledge of noise as described in Chapter 8.6 of "Foundations of the
Theory of Learning Systems" by Y. Tsypkin,
Academic Press, New York (1973). In such case a-priori knowledge of the noise is obtained via the identification process referred above and is not preassumed.
When both noise and information parameters are identified, the gain adjustment of a bandpass filter will be according to the signal-to-noise ratio in the bandpass to approximate a Kalman filter as described in page 35, Transactions ASME,
Journal of Basic Engineering (1960). In such a case, filtering will be facilitated even if the noise is white over the whole range of frequencies of interest.
The bandpass filters used may be digital filters, or discrete analog filters employing microelectronic tapped analog delays, or conventional analog filters. The combination of analog and logic hardware can also be utilized.
Some examples of apparatus in accordance with the invention will now be described with reference to the accompanying drawings, wherein: Fig. 1 is a block diagram of one example of an adaptive filter;
Fig. 2 is a frequency diagram showing the different bands into which an input signal containing information and noise is divided;
Fig. 3A is a plot showing the time-variation of a running average of the absolute value of a signal in a band in the absence of noise;
Fig. 3B is a plot like Fig. 3A showing the vertical shifting of the running average of the absolute value of a signal when near-stationary noise is present;
Fig. 4 is a block diagram of a second example of an adaptive filter;
Fig. 5 is a detail of one of a typical notch-filter of the noise-elimination circuit shown in Fig. 4;
Fig. 6 is a block diagram of a threshold logic circuit shown in Fig; 4 and
Fig. 7 is a block diagram of a hearing aid incorporating the first embodiment of the adaptive filter but providing for suppression of acoustic feedback.
Referring now to the drawings, reference numeral 100 designates a first example of an adaptive filter. Filter 100 comprises noiseanalysis circuit 102 and noise-reduction circuit 104. Each of circuits 102 and 104 comprises a plurality of bandpass filters that divide the input signal into a plurality of frequency sensitive bands 1 , 2, . . N, the bands in each of the circuits having a one-to-one correspondence. An input signal comprising both information and noise of the type described, is applied via line 11 in parallel to each bandpass filter. Each of the bands in circuit 102 contains a bandpass filter 130, a detector 140, and a smoothing circuit 1 50 serially connected together. Each component in a band is designated by a reference numeral associated with the component followed by a hyphen and a number designating the band.For example, bandpass filter BPF 1 of band 1 of circuit 102 is designated 130-1.
Each of the bands 1, 2,... N, covers a unique band within the range of frequencies associated with the information. This is illustrated in Fig. 2 which shows an information spectrum as extending from frequency f0 to fn and being divided into N bands that correspond to the N bands of the filters of circuit 1 02. If the information is speech, the spectrum could extend from 50 Hz to 6500 Hz in N steps, the upper limit having been chosen to correspond to the upper cut-off of hearing aids in current use. If the information is music, the spectrum would be substantially larger. Ideally each bandpass should have sharply defined cut-offs of infinite slope, but it is sufficient for the bandpass of the filter to approximate the ideal situation illustrated in Fig.
2.
A detector 1 40 in a band downstream of a bandpass filter 130 converts the signal passed thereby into the absolute value of such signal, or a function thereof, as for example, the RMS value of the signal. The absolute value signal is passed through an averaging circuit 1 50 that smooths the signal, preferably by producing a running average of the absolute value. The resultant smoothed signal is an analog form, and is converted to digital form by analog-to-digital converter 1 60.
The output of each band is then applied to microprocessor 170 of circuit 1 02. The microprocessor can be programmed to process the applied data for the purpose of detecting the presence of noise in the input signal, identifying the noise or noise and information parameters if noise is detected, and adjusting the gain of each of the bandpass filters 1 80 of circuit 104 in a way that reduces the noise in the output of this circuit.
Alternatively, the outputs of bandpass filters 1 30 could be applied directly to the microprocessors 1 70 through an ADC unit. In such case, the microprocessor would be programmed to compute the absolute value, smooth the same, and then examine the data to determine whether noise is present in the input signal before identifying the parameters, etc.
The presence of noise in the input signal is assumed when the smoothed signal in any band is such that a threshold is exceeded by all or by a predetermined percentage of successive minima of the smoothed signal. For example, the criterion for deciding if noise is present may be whether two out of three successive minima exceed a threshold in a given band. This test for noise is easily carried out by detecting minima in the smoothed signal in a band, storing three successive minima and comparing the value of each with a threshold. If less than two exceed the threshold, the next minimum produced replaces the fist minimum stored in this group, and a new threshold test is made.Reference can be made to
Fig. 3A which represents, in idealized form, the output in one band of averaging circuit 1 50, i.e., a smoothed signal whose variation with time is due solely to information in the input signal. As indicated in Fig. 3B, noise begins between intervals 10 and 12 on the time scale.
Microprocessor 1 70 might sample the smoothed signal at the times indicated in the time scale in
Fig. 3B with the result that by interval 15, three minima would have occurred, namely the minima at the second, ninth, and fifteenth time interval.At that instant, three registers (not shown) would contain the values AV(2), AV(9) and AV(15) as indicated. A comparison of these values with the threshold would show that only one out of the three minima exceeds the threshold. Under the critierion established above, a decision would be reached at the 15th interval that no noise is present in the input. When the next minima occurs at interval 20, the three registers in the microprocessor would contain the minima AV(9),
AV(15) and AV(20) associated with the ninth, fifteenth, and twentieth intervals.When the test is made at the twentieth interval, two out of the three amplitudes would exceed the threshold with the result that the microprocessor would conclude that noise was present in the input signal.
Alternative to the above criterion, a value of the inverse signal-to-noise ratio (SNR-1) in a band in excess of a threshold can be used as a test for the presence of noise in the input signal. In order to compute SNR-1, the smoothed signal in a band is sampled at intervals over which the information parameters in the input signals are substantially constant (e.g., at intervals of 10-30 milliseconds for speech). The SNR-' in a sampling intervalsis computed by taking the ratio of the last stored lowest minimum of the smoothed signal to the difference between the sampled value of the smoothed signal and such minimum. Fig. 3 illustrates computation of (SNR-1)12 which is given by AV(9)/AV(1 2)-AV(9))=a/b.
Once a decision is made that noise is present in the input signal, either noise parameters alone or noise and information parameters in each band are identified in accordance with the minimization process to be followed in controlling the filters of circuit 102. If noise parameters are to be identified, they may be in the form of predetermined functions of the lowest one of a predetermined number of successive minima of a smoothed signal in a band. Updating occurs in accordance with changes in the minima.
Alternatively, such parameters may be in the form of predetermined function of a more recent minimum of a succession of minima if such recent minimum does not exceed the previous lowest minimum by more than a given percentage (e.g., 25%). In either event, the identified parameters are used to adjust the gain elements of the bandpass filters 1 80 of circuit 102 so that the summed outputs of filters 180, carried out in summing circuit 200, produces an output signal in which the noise is minimized under a continuous minimization criterion thereby enhancing intelligibility of the information in the input signal. A suitable gain-to-parameter relationship that will minimize noise is that which approximates a Weiner filter with a-priori knowledge of the noise.In this case, however, apriori knowledge of the noise is obtained via the identification process described above and is not preassumed.
if the noise is expected to the white and to exist over the whole range of frequencies of interest, it is advantageous to identify the noise and information parameters in each band. In such case, microprocessor 1 70 would periodically compute the SNR in each band by computing the difference between the sampled smoothed signal in a band and the latest stored lowest minimum.
At the twelfth interval in Fig. 3, such difference is AV(12)-AV(9). The noise and speech parameters are then identified as continuous predetermined functions of such difference. The identified parameters are used to adjust the gain of the bandpass filters 1 80 of circuit 102 so that the summation of the outputs of filters 1 80 carried out in summing circuit 200 produces an output signal which the noise minimized under a continuous minimization criterion thereby enhancing intelligibility of the information in the output signal. A suitable gain-to-parameter relationship that will minimize noise is that which approximates a Kalman filter as described above.
In order to avoid long delays before a decision on the existence of noise is made, the initial nonzero minimum of the smoothed signal in a band will be decided upon as being noise-related and acted upon accordingly, and possibly wrongly, until the next successive minimum occurs. The original decision will be confined by the occurrence of other minima of value close to the original minimum within a relatively small predetermined number of minima. This will possibly result in a short period of improper filtering but it is a small price to pay for fast tracking and filtering of noise.
Microprocessor 170 shown in Fig. 1 has three functions, namely the detection of noise, the identification of the parameters of the noise, and providing gain adjustment information for the filters of circuit 102. These functions can be carried out as indicated above, by suitably programming a general purpose microprocessor or by hard wiring a system. The latter is illustrated in Fig. 1 wherein circuit 1 70 includes minima detector and storage means 106, comparator 107, parameter identifier 108 and parameter-togain converter 109. Means 106 comprises logic and control circuitry for sampling the smoothed signals in each band, for determining the existence of minima, and storing the same for comparison purposes.Alternatively, means 106 is capable of computing SN R-l. Threshold comparison is carried out by comparator 107 whose output 110, with respect to each band, is indicative of noise in any band. If desired, the system can be organized so that noise is assumed only if the threshold is exceeded in each band.
Parameter identifier 108 is schematically shown as a separate block, but it should be understood that if noise-only parameters are of interest, they will be function of the lowest minima as determined by means 106. If noise and information parameters are of interest, they will be functions of SNR-' (namely SNR) which can be computed by means 106. Finally, converter 109 establishes the values of the gain elements of noise-reduction circuit 104 in accordance with
Weiner or Kalman filter theory depending on the nature of the parameters identified.
In general, the noise is tracked on the basis that noise parameters change at a rate considerably slower than those of speech, the latter changing considerably over intervals of about 1 5 milliseconds. Hence, the data for computing noise parameters is based on the lowest minimum in a sequence of say two or three successive minima of the absolute values of a smoothed output in a band, where smoothing is sufficient to eliminate variations in frequencies above say 30 Hs. Furthermore, if three successive minima are considered, and one has a value of say twice the previous one, then the next minima is considered to represent a drop in speech.If all three later minima are, however, considerably higher than the initial minima, this serves to indicate that noise parameters have changed, the comparison of the three successive minima serving as a test of the stationary nature of the noise. If only a gradual change of a few percent is observed in the value of successive minima, this serves to indicate a slow drift in noise parameters within the framework of quasi-stationarity, and the noise parameters are then updated accordingly. Speech parameters, however, are updated at every 1 0-30 milliseconds and no stationarity test is performed on them.
The apparatus is particularly well suited for use in hearing aids to facilitate transmission of intelligence in noisy environments. However, the ability of the above described embodiment to selectively filter noise in a given band without affecting transmission other bands permits acoustic feedback to be compensated for without significantly reducing the performance characteristics of the filter. Because acoustic feedback in a hearing aid occurs in a particular band, identification of the band and the provision of suitable gain reduction in that band will eliminate the ringing problem.
Ringing in a hearing aid is dependent on the hearing aid system itself, and thus cannot be treated as environmental noise. Indeed, with respect to ringing, an adaptive filter incorporated into a hearing aid is, itself, a part of the feedback loop that generates the noise. Consequently, elimination of ringing requires the adaptive filter loop to be broken such that the frequency band in which ringing occurs can be identified. During identification, the adaptive nature of the filter is suppressed. Once the frequency band in which ringing takes place is identified, the gain of the filter in the identified band is reduced to eliminate ringing. Thereafter, the adaptive nature of the filter is resumed while maintaining the gain in the band in which ringing was detected at a level below which ringing can occur.
In practice, Fig. 7 shows one approach to solving this problem. Hearing aid 201 comprises a microphone 202 driving a conventional amplifier 204 whose output is applied to electronic switch 206 whose state establishes the mode of noiseanalysis circuit 208 that is similar to circuit 102 of the first embodiment. The state of switch 206 is controlled by a delay circuit 209 which is responsive to closure of power switch 210 when the hearing aid is turned on for maintaining switch 206 in an initial state that places circuit 209 in a feedback testing mode. In such mode, which may last about two seconds, circuit 208 is effective to identify any band whose output, during the test interval, is constantly well above the outputs of any of the other bands by more than a predetermined value.When such a band is identified, the gain of the corresponding filter in noise-reduction circuit 211 is reduced to a level that extinguishes the ringing. Such level is remembered by circuit 208 when switch 206 changes state and circuit 208 is placed in its steady-state adaptive filtering mode. In its latter state, circuit 208 adaptively filters noise pickedup by microphone 202 from the output of speaker 212, but the gain in the band in which ringing had been detected is maintained at the remembered level.
A second embodiment of an adaptive filter 10 is shown in Fig. 4. Filter 10 comprise a noise analysis circuit and a noise-reduction circuit 1 2.
Each of these circuits comprises a plurality of bandpass filters that divide the input signal into a plurality of frequency sensitive bands 1, 2,... N, the bands in each of the circuits having a one-toone correspondence. An input signal, comprising information and noise of the type described is applied via line 11 in parallel to each bandpass filter. Each of the bands in circuit 1 2A contains a bandpass filter 1 3, a detector 14 and a threshold logic circuit 1 5 serially connected together. Each component in a band is designated by a reference numeral associated with the component followed by a hyphen and a number designating the band.
For example, the bandpass filter of channel 1 is designated 13-1.
A detector 1 4 in a band downstream of a bandpass filter converts the signal passed thereby into a representation of the absolute value of such signal or a function thereof. The output of each absolute value detector is passed through a threshold logic circuit where it is sampled by a sampling gate under the control of a clock pulse.
Preferably, the threshoid logic circuit stores representations of the minima of the sampled absolute values over a predetermined time interval, but may also compute the signal-to-noise ratio as described above. A decision on whether the input signal contains noise can be based on either of the two criteria described above, namely whether a predetermined percentage of successive minima of the smoothed absolute value in a band exceeds a threshold, or whether
SNR-' exceeds a threshold. As shown in Fig. 6, a threshold logic circuit in a band includes a sampler circuit 1 6 for sampling the output of the absolute value detector of the band and converting its continuous output into discrete values which may be operated on using digital techniques, and a threshold circuit 1 6A for detecting minima in the discrete output.Circuit 1 6A produces a control pulse each time the threshold is exceeded, such pulses being counted by decision circuit 16B.
If the first noise detection criterion is utilized, then a predetermined number of successive minima will exceed the threshold of circuit 16A; and the output of circuit 1 6B is indicative of the presence of noise of the type described in the input signal. If the converse is true, the output of channel 1 6B is indicative of the absence of noise in the input signal. In the event SNR-' is computed and used as the criterion for determining whether noise is present, circuit 1 6B would indicate noise is present in a channel if
SNR-', computed in the manner described above, exceeds a threshold. Regardless of the arrangement employed to detect the presence of noise, circuit 1 6B provides an output to parameter storage means 1 7 shown in Fig. 4 identifying a band as one containing noise.
Selectively adjustable noise reduction circuit 1 2 includes a serial array of bandpass filters 1 8 to which the input signal is applied in parallel with its application to the noise-analysis circuit. The filters 1 8 can be variable gain bandpass filters as in the case of the first embodiment, or they may be in the form of notch filters centered on the band in which they are located. Filter 1 8-1 is thus associated with band 1 of the speech spectrum, and with bandpass filter 13-1 of band 1 of the noise-analysis circuit.
Parameter storage means 17 supplies control signals via lines 1 9 to circuit 1 2 for the purpose of selectively activating those notch filter means associated with bands containing noise of the type described. The input signal passes serially through those notch filter means associated with bands that have been identified as containing noise, and bypasses the remaining notch filter means. As shown in Fig. 1, an input signal that bypasses a given notch filter means appears in line 20 associated therewith while an input signal that is filtered by the notch filter appears in line 21. The selection as to whether a given filter means will be bypassed by the input signal, or will filter the same, is determined by the control signal applied to the notch filter means.
Fig. 5 shows a suitable circuit for a typical notch filter means. The input to the notch filter means may appear either in line 22 o;in line 23 depending upon whether the upstream notch filter means has been activated by a control signal applied thereto. In the case of notch filter 18-1, one or the other of lines 22, 23 is not used and the input signal from line 11 is applied to the other. In the absence of an enable signal applied to gate control line 25 of a notch filter means, an input signal in either of lines 22, 23 bypasses notch filter 26 and appears at the bypass line 20.
If the upstream notch filter means had not been activated, the signal appearing in line 22 passes through gate 27 by reason of its normally closed state. If the upstream notch filter means had been activated, a signal appearing in line.23 passes through normally closed gate 28 and thus appears in line 20.
A notch filter means is activated when one of the control lines 1 9 supplies an enabling gate signal to the gate control 25 thereby changing the normal states of gate 27-30. Specifically, gates 27 and 28 are opened by a gate signal and gates 29 and 30 are closed. Thus, a signal in either of lines 22 or 23 will be applied to notch filter 26 so that an output appears on filter line 21. In this manner, the input signal in line 11 is routed through the notch filters of the activated notch filter means and bypassed around the notch filters of those filter means that are not activated.
In operation, the presence of noise of the type described in the input signal in line 11 is detected by one or more of the threshold logic circuit 1 5 in the N frequency selective channels. The noise or noise and information parameters of such noise can be identified in each band in the manner previously described. Moreover, these parameters are identified at any time, including during a speech interval, without assuming a-prior knowledge of the particulars of the noise or even of its existence. The output signals of the threshold logic circuits specify those notch filters of the notch filter means through which the input signal will be applied for the purpose of adapting the selectively adjustable filter to a condition that filters most of the noise from the input signal. The degree to which the noise is filtered out depends on how close the noise frequencies are to the center-frequencies of the various notch filters and on the SNR levels of the corresponding bandpass filters. However, since many bands are contemplated, it is possible to insure adequate filtering by providing for a sufficient number of bands. Instead of using the control signals in line 1 9 for completely inserting or removing filters 18, the technique described above in connection with the first embodiment of an adaptive filter can be used, namely, adjusting the gain, and possibly the notch frequency, of the filters.
The bandpass and notch filters shown in the drawings may be digital filters or discrete analog filters employing microelectronic tapped analog delays, or analog filters. The latter two approaches facilitate very fast operation and eliminate the need for analog-to-digital conversion circuitry. Furthermore, while Wiener and Kalman filter theory is described above in connection with the processes for minimizing noise, heuristic minimization criteria could also be used.
Claims (3)
1. An adaptive filter for adaptively filtering near stationary, relatively long duration noise, as hereinbefore defined, in an input signal containing information such as speech whose parameters change more slowly than the noise parameters, the filter comprising a noise analysis circuit for detecting the presence of noise of the type described in the input signal and for identifying the parameters representing the characteristics of at least the noise independently of whether information is present in the input signal; and a noise reduction circuit having gain elements controlled by the identified parameters for minimising noise in the output of the circuit.
2. An adaptive filter according to claim 1, wherein the noise analysis circuit includes means for dividing the input signal into a first set of signals limited to different frequency bands; means for producing a function of the absolute value of the signal in each band and for smoothing the function to provide a smoothed signal; means for sampling the smoothed signal in a band and taking the ratio of the latest stored lowest minimum of the smoothed signal in a band to the difference between the sampled value of the smoothed signal and the minimum; and means for comparing the ratio in each band with a threshold for detecting the presence of noise in the input signal.
3. An adaptive filter according to claim 1, wherein the noise analysis circuit includes means for dividing the input signal into a first set of signals limited to different frequency bands; means for producing a function of the absolute value of the signal in each band and for smoothing the function to produce a smoothed signal; means for storing successive minima of the smoothed signal in a band; and means for comparing the amplitude of the stored minima in a band with a threshold for detecting the presence of noise in the input signal.
3. An adaptive filter according to claim 1 wherein the noise analysis circuit includes means for dividing the input signal into a first set of signals limited to different frequency bands; means for dividing the input signal into a first set of signals limited to different frequency bands; means for producing a function of the absolute value of the signal in each band and for smoothing the function to produce a smoothed signal; means for storing successive minima of the smoothed signal in a band; and means for comparing the amplitude of the stored minima in a band with a threshold for detecting the presence of noise in the input signal.
4. An adaptive filter according to claim 3, wherein the noise analysis circuit includes means responsive to the lowest one of a number of successive minima of the smoothed signal in a band for forming a function representative of a frequency sensitive noise parameter that identifies the same.
5. An adaptive filter according to claim 3, wherein the noise analysis circuit includes means responsive to a minimum other than the lowest of a succession of minima in a band if such other minimum does not exceed the lowest minimum by more than a given percentage for forming a function representative of a frequency sensitive noise parameter that identifies the same.
6. An adaptive filter according to claim 3, wherein the noise analysis circuit includes means responsive to the difference between samples of the smoothed signal in a band and the most recent noise-related minimum of a time sequence of minima of the smoothed signal for forming a function representative of a frequency sensitive noise and information parameter that identifies the same.
7. An adaptive filter according to claim 4, or claim 6, further including means for dividing the input signal into a second set of signals limited to the same bands as the first set; and means for controlling the gain of each band in accordance with the identified parameters.
8. An adaptive filter according to claim 7, when dependant on claim 4, wherein the gain-toparameter relationship approximates a Weiner filter with a-priori knowledge of noise obtained via the identification process.
9. An adaptive filter according to claim 7, when dependant on claim 6, wherein the gain-toparameter relationship approximates a Kalman filter with a-priori knowledge of the noise and information obtained via the indentification process.
1 0. A filter substantially as described with reference to any one of the examples illustrated in the accompanying drawings.
11. An assembly comprising a hearing aid including an adaptive filter according to any of the preceding claims; a microphone for converting sound into a representation of the input signal to the filter; a speaker for converting the output of the filter into sound; and means for suppressing acoustic feedback.
12. An assembly according to claim 11, wherein the noise analysis circuit of the filter includes means for dividing the input signal into a first and a second set of signals limited to the same frequency bands; means for testing for the presence of ringing in each band of the first set during operation of the hearing aid and identifying the band containing ringing without affecting the second set of signals; means for adjusting the gain in each band of the second set; and means
responsive to identification of a band containing ringing for reducing the gain in the band of the second set which contains ringing to a level that
reduces the ringing.
New claims or amendments to claims filed on 6
Sep 1983
Superseded claims 1 and 3
New or amended claims:
1. An adaptive filter for adaptively filtering near stationary, relatively long duration noise, as hereinbefore defined, in an input signal containing information such as speech, the filter comprising a noise analysis circuit for detecting the presence of noise of the type described in the input signal and for identifying the parameters representing the characteristics of at least the noise independently of whether information is present in the input signal; and a noise reduction circuit having gain elements controlled by the identified parameters for minimising noise in the output of the noise analysis circuit.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
GB08318836A GB2126851B (en) | 1979-08-17 | 1983-07-12 | Adaptive filter |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
GB7928813A GB2057228B (en) | 1979-08-17 | 1979-08-17 | Method and means for adaptively filtering near-stationary noise from an information bearing signal |
GB08318836A GB2126851B (en) | 1979-08-17 | 1983-07-12 | Adaptive filter |
Publications (3)
Publication Number | Publication Date |
---|---|
GB8318836D0 GB8318836D0 (en) | 1983-08-10 |
GB2126851A true GB2126851A (en) | 1984-03-28 |
GB2126851B GB2126851B (en) | 1984-08-15 |
Family
ID=26272615
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
GB08318836A Expired GB2126851B (en) | 1979-08-17 | 1983-07-12 | Adaptive filter |
Country Status (1)
Country | Link |
---|---|
GB (1) | GB2126851B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0491063A1 (en) * | 1990-12-17 | 1992-06-24 | Siemens Aktiengesellschaft | Adaptive balance filter |
GB2433688A (en) * | 2005-12-22 | 2007-06-27 | Red Lion 49 Ltd | Amplifying a filtered audio signal based on a control signal from filtered audio signal |
US9913051B2 (en) | 2011-11-21 | 2018-03-06 | Sivantos Pte. Ltd. | Hearing apparatus with a facility for reducing a microphone noise and method for reducing microphone noise |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8712076B2 (en) | 2012-02-08 | 2014-04-29 | Dolby Laboratories Licensing Corporation | Post-processing including median filtering of noise suppression gains |
US9173025B2 (en) | 2012-02-08 | 2015-10-27 | Dolby Laboratories Licensing Corporation | Combined suppression of noise, echo, and out-of-location signals |
-
1983
- 1983-07-12 GB GB08318836A patent/GB2126851B/en not_active Expired
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0491063A1 (en) * | 1990-12-17 | 1992-06-24 | Siemens Aktiengesellschaft | Adaptive balance filter |
GB2433688A (en) * | 2005-12-22 | 2007-06-27 | Red Lion 49 Ltd | Amplifying a filtered audio signal based on a control signal from filtered audio signal |
GB2433686A (en) * | 2005-12-22 | 2007-06-27 | Red Lion 49 Ltd | Amplification of filtered audio input signal |
GB2433688B (en) * | 2005-12-22 | 2011-08-10 | Red Lion 49 Ltd | Processing an audio input signal to produce a processed audio output signal |
US8094837B2 (en) | 2005-12-22 | 2012-01-10 | Red Lion 49 Limited | Processing an audio input signal to produce a processed audio output signal |
US8666093B2 (en) | 2005-12-22 | 2014-03-04 | Red Lion 49 Limited | Processing an audio input signal to produce a processed audio output signal |
US9913051B2 (en) | 2011-11-21 | 2018-03-06 | Sivantos Pte. Ltd. | Hearing apparatus with a facility for reducing a microphone noise and method for reducing microphone noise |
US10966032B2 (en) | 2011-11-21 | 2021-03-30 | Sivantos Pte. Ltd. | Hearing apparatus with a facility for reducing a microphone noise and method for reducing microphone noise |
Also Published As
Publication number | Publication date |
---|---|
GB8318836D0 (en) | 1983-08-10 |
GB2126851B (en) | 1984-08-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US4185168A (en) | Method and means for adaptively filtering near-stationary noise from an information bearing signal | |
EP0077574B1 (en) | Speech recognition system for an automotive vehicle | |
US7155385B2 (en) | Automatic gain control for adjusting gain during non-speech portions | |
KR100335162B1 (en) | Noise reduction method of noise signal and noise section detection method | |
JP4187795B2 (en) | Method for reducing speech signal impairment | |
US6820053B1 (en) | Method and apparatus for suppressing audible noise in speech transmission | |
US6038532A (en) | Signal processing device for cancelling noise in a signal | |
EP0763813B1 (en) | Speech signal processing apparatus for detecting a speech signal from a noisy speech signal | |
US4025721A (en) | Method of and means for adaptively filtering near-stationary noise from speech | |
US5097510A (en) | Artificial intelligence pattern-recognition-based noise reduction system for speech processing | |
US8311819B2 (en) | System for detecting speech with background voice estimates and noise estimates | |
US6289309B1 (en) | Noise spectrum tracking for speech enhancement | |
US5479517A (en) | Method of estimating delay in noise-affected voice channels | |
JP3423906B2 (en) | Voice operation characteristic detection device and detection method | |
EP0459364A1 (en) | Noise signal prediction system | |
US4597098A (en) | Speech recognition system in a variable noise environment | |
EP2244254A1 (en) | Ambient noise compensation system robust to high excitation noise | |
RU96102854A (en) | METHOD AND DEVICE FOR NOISE REDUCTION IN SPEECH SIGNAL | |
KR20060094078A (en) | Voice activity detection with adaptive noise floor tracking | |
JPH0213999A (en) | Voice signal detection method and apparatus | |
US8199928B2 (en) | System for processing an acoustic input signal to provide an output signal with reduced noise | |
US5768392A (en) | Blind adaptive filtering of unknown signals in unknown noise in quasi-closed loop system | |
JPS59105696A (en) | Voice pause recognition | |
GB2126851A (en) | Adaptive filter | |
KR100361235B1 (en) | Signal Source Characterization System |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PCNP | Patent ceased through non-payment of renewal fee |