CN100589184C - Subtractive cancellation of harmonic noise - Google Patents
Subtractive cancellation of harmonic noise Download PDFInfo
- Publication number
- CN100589184C CN100589184C CN200510074622A CN200510074622A CN100589184C CN 100589184 C CN100589184 C CN 100589184C CN 200510074622 A CN200510074622 A CN 200510074622A CN 200510074622 A CN200510074622 A CN 200510074622A CN 100589184 C CN100589184 C CN 100589184C
- Authority
- CN
- China
- Prior art keywords
- frequency
- signal
- disturbance
- estimation
- sinusoidal
- 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.)
- Expired - Fee Related
Links
Images
Landscapes
- Filters That Use Time-Delay Elements (AREA)
- Soundproofing, Sound Blocking, And Sound Damping (AREA)
Abstract
A common problem in audio processing is that a useful signal is disturbed by one or more sinusoidal noises that should be suppressed. One embodiment of the invention provides a method of canceling a sinusoidal disturbance of unknown frequency in a disturbed useful signal. The method comprises the steps of estimating parameters of the sinusoidal disturbance including amplitude, phase and frequency;generating a reference signal on the basis of the estimated parameters; and subtracting the reference signal from the disturbed useful signal. The estimation is performed by an Extended Kalman filter.
Description
Technical field
Present invention relates in general to suppress the field of noise, more specifically, relate to a kind of method of eliminating the additional sinusoidal perturbation of the unknown frequency in institute's attention signal.This method concentrates on the quality that improves vision signal.Yet the present invention is not limited to field of acoustics, that is, it can be applied to the signal of pressure transducer.
Background technology
Common issue with in the Audio Processing is the interference that information carrying signal is subjected to one or more sinusoidal signal.The classic method that suppresses these undesired signals is to use the fixed trap wave filter that frequency is transferred to this sinusoidal interference frequency, as at " Halbleiter-Schaltungstechnik " (by Ulrich Tietzeand Christoph Schenk, Springer, 12th edition, 2002) described in.
Decline slightly only takes place in the quality of attention signal in order to make, and requires the trap of wave filter very steep, for good inhibition, need accurately know interfering frequency.If not so, then the conventional method of notch filter is no longer suitable, and the adaptive approach that proposes in must using in " Adaptive IIR Filtering inSignal Processing and Contrl " (by Philip A.Regalia Marcel Dekker, 1994).This wave filter is synchronous with the main sinusoidal interference that contains highest frequency, and suppresses this interference fully.In addition, this wave filter can be followed the trail of the less variation of being comply with in the time of this interfering frequency.Yet a major defect of this method is: it does not preserve the spectrum content of the beared information at trap frequency place.Can't clearly separate two sine waves (represent noise for, another represents useful information) thus.
When being considered as elimination to disturbance, the inhibition that offset of sinusoidal is disturbed to overcome the above problems.Generate artificial reference signal and from the noise information carrying signal, deduct this signal.The quality of the estimated value of the sine parameter that is used for reference signal is depended in this inhibition at present.
In case found good estimation, then can slow down and estimate handle or it is stopped fully, so that estimator can't be followed the tracks of the amplitude that caused by institute's attention signal and the variation of phase place.As long as the sinusoidal interference parameter keeps constant for the time, just can preserve this spectral content.If these parameters change, then no longer preserve this spectral content, and force and restart conventional estimation procedure.The frequency that the method for prior art hypothesis will be eliminated is known, and most of method is for the continuous parameter estimated service life gradient descending method of amplitude and phase place, for example "
Mitmodellbasierten
F ü r Freisprecheinrichtungen in Kraftfahrzeugen " (by Henning Puder, PhD Thesis, Technische
Darmstadt, 2003).For processes voice signals utilizes the decline step-length to control estimation for the disturbance sine parameter, just and this estimation only when speech pause, be activated.Like this, significantly reduced inhibition for the useful frequency content in the phonological component.
Summary of the invention
Some the purpose of this invention is to provide a kind of improved noise cancellation technique more than considering, it can also be applied to the interfering frequency condition of unknown.
Feature by independent claims realizes described purpose.In independent claims, defined favorable characteristics.
The present invention has removed other sinusoidal interference basically by compensation technique from the voice signal that is subjected to disturbance.Basic skills is to use the inphase/orthogonal model of sinusoidal interference.
The method estimation that is proposed is also followed the tracks of following parameter: in-phase amplitude, orthogonal amplitude and frequency that each disturbs.Kalman filter by expansion is recursively carried out this estimation.According to these three parameters, come sinusoidal interference in the compensate for disturbances signal by generating reference signal and from undesired signal, deducting this reference signal.
Kalman filter by expansion is carried out the estimation of three unknown sinusoidal interference parameters successively.Be similar to adaptive notch filter, this wave filter converges on the strongest frequency of power, and estimates its parameter.By selecting different value, come parameter estimation procedure is controlled for the noise of equipment covariance in the measured value of being supposed and the Kalman's system.The higher value of measuring in the covariance has for example been fixed estimated value and reference signal.The advantage of method proposed by the invention is, need not to know interfering frequency, and opposite with adaptive notch filter, not erasure signal information.
Suppressing under the situation of noise of motor, can be identified for each value of the variance of the initialization of Kalman filter and signal and interference by additional sensor such as the revolution meter of motor.Can also determine these values by learning process, in this process, identify possible disturbance/interference/noise and their character.The value of Que Dinging not is the explicit value of interfering frequency thus, and is its estimated value, and this is useful for accelerator card Thalmann filter self-adaptation and raising estimated accuracy.
In addition, in filtration treatment, independently measure equation and easily integrate (intergrate) initialization continuous sensor information afterwards by interpolation.Can finish the sensor associating of revolution meter and other device thus.
A kind of method of sinusoidal perturbation of the unknown frequency that is used for eliminating the useful signal that is subjected to disturbance is provided according to a first aspect of the invention.Thus, this method may further comprise the steps: estimate three parameters of sinusoidal perturbation, i.e. amplitude, phase place and frequency; Generate reference signal according to estimated parameter, and from the useful signal that is subjected to disturbance, deduct reference signal.
Can utilize the value of additional sensor and/or the estimated value of learning process offset of sinusoidal excitation parameter to carry out initialization.
Particularly, can eliminate a plurality of sinusoidal perturbations by repeating this method continuously.
Before estimating step, the useful signal that is subjected to disturbance is carried out bandpass filtering.
Before being applied to each wave band, this method can utilize a plurality of bandpass filter will be subjected to the useful signal of disturbance to be decomposed into a plurality of frequency ranges thus.
In addition, can eliminate the given sinusoidal perturbation in first frequency range, and can eliminate the given sinusoidal perturbation in second frequency range by the reference signal that generates for the given sinusoidal perturbation of eliminating in first frequency range.
Can the given sinusoidal perturbation in second frequency range be eliminated the ratio of response of first band frequency and the response of second band frequency by adjusting the reference signal that generates for the given sinusoidal perturbation of eliminating in first frequency range.
Can carry out this estimation by the Kalman filter of expansion.
In addition, can adjust the degree of confidence of the initialization value in the estimating step.
Can also adjust degree of confidence by the error covariance matrix of controlling the Kalman filter of expanding then.
Can selection of time ground and especially according to speech activity measurement carry out this method.
Can carry out filtering to the estimated useful signal that is obtained according to Ephraim and Malah method.
According to a further aspect in the invention, propose a kind of computer software program product, when operating in it on calculation element, realized preceding method.
According to a further aspect in the invention, provide a kind of system that is used for eliminating in the sinusoidal perturbation of the unknown frequency of the information carrying signal that is subjected to disturbance, wherein calculation element is carried out said method.
Description of drawings
According to following detail specifications and claims in conjunction with the accompanying drawings, other advantage of the present invention will become clear with using.Wherein,
Fig. 1 represents to eliminate noise in the signal that is subjected to disturbance according to of the present invention by adding reference signal,
Fig. 2 represents recurrence Kalman algorithm for estimating, and
Fig. 3 represents recurrence spreading kalman algorithm for estimating.
Embodiment
Compensation method
Below with reference to Fig. 1 entire compensation method of the present invention is described, this method proposes: eliminate noise in the signal that is subjected to disturbance by adding reference noise.
As can be seen from Figure 1, the following parameter for each interference is also followed the tracks of in the method estimation (2) of the present invention's proposition: in-phase amplitude, orthogonal amplitude and frequency.Kalman filter by expansion comes recurrence to carry out this estimation.Subsequently,, generate (4) reference signal (5), and from the signal (1) that is subjected to disturbance, deduct (6) this signal, be subjected to the sinusoidal perturbation (9) in the signal (1) of disturbance with compensation according to three estimated parameters (3).
Employed reference signal is the artificial signal (5) that generates according to noise model (4)
This artificial signal indication is superimposed upon the estimated value of the actual turbulent noise (9) on information carrying signal (8) s (n).Carry out the estimation (2) of described reference signal indirectly by determining following model parameter:
By from whole signal (1) y (n) that is subjected to disturbance, deducting (6) artificial model signal (5)
Suppress noise (9):
Wherein e (n) is the error signal after the noise compensation of moment n,
S (n) is the useful signal at moment n,
V (n) is the interference noise of moment n,
Y (n) is the useful signal that is subjected to extra disturbance of moment n.
The suitable model of handling the sine-wave oscillation compensation is an inphase/orthogonal model used in the present invention.In this model, can be by following three parameter (θ
1, θ
2And θ
3) the general sinusoidal signal v (n) according to formula (3) described,
θ
1=Acos φ formula 4a
θ
2=Asin φ formula 4b
These three parameters are represented in-phase component, quadrature component and normalized frequency respectively.
The generation of reference signal is described by following formula:
V (n, θ)=θ
1Cos (2 π θ
3N)-θ
2Sin (2 π θ
3N) formula 5
This method has been eliminated the shortcoming of notch filter basically.This makes:
1, the determined vibration that decays targetedly, rather than delete them fully.Can keep the constant and lasting vibration of useful signal thus.
2, utilize the constant estimation of model parameter according to input signal and last estimated value
Temporarily follow the tracks of the variation of interfering frequency:
The result who obtains by described method depends on the precision of estimator (2) and the probability of distinguishing useful signal (8) and noise signal (9).The little evaluated error of phase place or frequency may cause the mistake that subtracts each other between reference signal and the noise signal over time.So constant new estimated value (2) of absolute demand.To remain on low-levelly in order making to assess the cost, to the present invention proposes the use sequential grammar.
Kalman filter
Illustrate with reference to Fig. 2 and Fig. 3 how the present invention utilizes estimation of the order method, i.e. Kalman filter below.
In order to calculate current estimated value
Kalman filter only need be subjected to current sample value y (n)=s (the n)+v (n) of the signal of disturbance, the last estimation of parameter
And the information that adopts error covariance matrix M (n-1|n-1) form about described estimated accuracy.In addition, the favorable characteristics of wave filter is that it provides the optimum linear estimated result to the parameter θ of linear change (n) in time, can be referring to " Fundamentals of Statistical Signal Processing-Estimation Theory ", (Steven M.Kay, Signal Processing Series, Prentice Hall, 1993).Optimum estimate value representation Kalman filter makes expectation quadrature error (that is linear least squares error (LMMSE)) minimum of all Linear Estimation devices.
The following describes the elimination that deducts that how general Karman equation is adjusted into according to harmonic noise of the present invention.
Because standard method needs linear dynamic model, so at first suppose the 3rd parameter (promptly
) known.Describe to use according to the Kalman filter of expansion of the present invention with the lower part in, existing equation is improved, and has increased Frequency Estimation.
The parameter θ that estimates (n) is the state variable of system.Simulate these parameters over time by linear stochaastic system.
θ (n)=A θ (n-1)+Bu (n), n 〉=0 formula 7
θ wherein
1(n) and θ
2(n) the current homophase and the quadrature component of regulation sinusoidal perturbation, and u (n) is the zero average two-dimentional white noise of normal distribution
(0, Q) formula 9 for u～N
Channel θ wherein
1(n) and θ
2(n) uncorrelated each other, and have identical variance
Formula 10
Can observe parameter θ (n) by noise signal (1) y (n) that is subjected to disturbance:
Wherein w (n) has represented the influence of voice signal (8) s (n) to noise signal (9) v (n):
" voice noise " w (n) can pass through average value mu
w(n) and its variances sigma
w 2(n) come descriptive statistics.Yet this is for description fully of its statistical property and insufficient, because the hypothesis of Gaussian distribution is for this voice signal and be false.As a result, aspect least mean-square error (MMSE), Kalman filter can not generate optimum, and the optimum value that is used for the Linear Estimation method (LMMSE) only is provided.Fig. 2 represents recurrence Kalman's algorithm for estimating of obtaining from above definition and hypothesis.
Initialization comprises the value of setting
And M (1|-1).Algorithm is from n=0.Advise the initial value as mean value and covariance in theory at moment n=-1 operation parameter θ.Because be difficult to these parameters are distributed statistics, (reasonable guess value 1|-1) is as initial value so the present invention proposes to use θ.By the M (1|-1) degree of confidence of the described initial value of decision.In order to estimate homophase or quadrature component, suppose to use [00]
TAs mean value.Utilize following error covariance matrix, possible estimation range is restricted hardly:
Formula 13
σ
2＝100
If for σ
2, select little a lot of value, then in section sometime, this algorithm can be sought " correctly " the parameter θ (n) in the initial value scope.If this algorithm does not find described parameter, then it slowly changes its " direction of search ".Filter table reveals extremely strong " deflection ".
Can control amplitude θ by covariance matrix Q
1(n) and θ
2(n) tracking.According to the present invention, matrix Q is a diagonal matrix:
So that following the independent of two amplitude components changes.According to the present invention, the appropriate value of ground unrest is σ
u 2=10
-13The feature that excessive value caused looks like the feature of notch filter.
The Kalman filter of expansion
Illustrate below with reference to Fig. 3 how the present invention utilizes the Kalman filter of expansion.
Utilize above-mentioned wave filter, tracking frequencies that can't be suitable changes.Can change this problem by the Kalman filter algorithm of representing among Fig. 2 is added the 3rd recurrence formula that is used for frequency.Then Kalman filter can be synchronous with the vibration with variable frequency, and can follow the tracks of and compensation change in time.Unfortunately, this improvement can't be carried out in the theoretical field of common Kalman, and this is non-linear in frequency range because of following observation formula:
y(n)＝θ
1cos(2πθ
3n)-θ
2sin(2πθ
3n)+w(n)
Formula 15
＝h(θ(n)，n)+w(n)
Still can utilize the estimation of the order formula of Kalman filter in any case.In fact, by adopting the Taylor series approximation value, can be with item h (θ (n), n) linearization.Thus, as described in the following formula in estimated value
Near research benchmark model h (θ, n):
Formula 15 becomes then:
Present described formula is linear, and only is following known terms with the difference of Kalman's model formation (being formula 11):
By the conversion y ' (n)=y (n)-z (n), obtain the initial prerequisite identical with the normal state Kalman filter.When using kalman filter method, obtain algorithm for estimating shown in Figure 3 (Kalman filter (EKF) that is called expansion).
Prediction steps (step 1 and step 2) remains unchanged.Have only number of parameters to increase by 1 to 3.Frequency has been added inphase/orthogonal component to parameter.Three other formula of Kalman filter algorithm (step 4b, 5b and 6b) demonstrate variation slightly.Carry out the formula use nonlinear model of the correction of prediction estimated value according to the value y (n) of new measurement
Predict the expectation measured value
(step 5b).Amplification/gain (step 4b) and evaluated error (step 6b) are used the first rank linearization
Must calculate this value for each new step.Similar with linear kalman filter, can not gain and error process in calculated off-line.In addition, wave filter loses its linear optimal characteristic because of linearization, and evaluated error M (n|n) can be translated as the first approximation of actual error.
Frequency sub-band decomposes
To illustrate that below the frequency sub-band of being carried out by the present invention decomposes.
Directly do not carry out according to inhibition of the present invention on the voice signal of disturbance (1) y (n) being subjected to.On the contrary, the present invention proposes: at first carry out frequency sub-band and decompose, this is the first step that deducts elimination of harmonic noise.Its function has been reproduced the nerve signal of people's cochlea and has been handled.Carry out squelch in higher neural rank then, and use the signal that filters by cochlea.
The model of expression good result is that the gamma that Patterson proposes is transferred (Gammatone) bank of filters.Relevant therewith, referring to the technical report of Malcom Slaney: " An Efficientimplementation of the Patterson Holdsworth auditory filter bank " (AppleComputer Inc, 1993).Described bank of filters comprises a plurality of eight different rank bandpass filter, and wherein these wave filters have different mutually bandwidth and different centre frequency distances.Limit bandwidth and distance or frequency range according to psychoacoustic analysis and overlap, and they increase and increase along with frequency.
Propose as the example that the cochlea of robot head is simulated: the described gamma tunable filter group of using a kind of pattern with 100 channels.In the different frequency range limited channel of bank of filters, realized that the noise of sinusoidal perturbation reduces.According to forcing frequency, must in more than one channel, carry out inhibition, this is owing to may have identical convergent response in the adjacent channel that overlaps.Then also must be in other channel the disturbance suppression frequency.This means with directly handling (being notch filter) and compare, need considerable odd word.On the other hand, compensation technique according to the present invention has benefited from the frequency sub-band decomposition.Separate approaching sinusoidal interference by decomposing.This bank of filters especially demonstrates low channel width for dark frequency, makes it to separate and has high-power sine-wave oscillation (that is the 100Hz of network buzzing, (humming) and 200Hz vibration).
Only on a channel, carry out estimation procedure.Expediently, selected channel is the channel that has the peak swing stroke for given original frequency.Then the fixed relationship between the transfer function of main channel and combined channel (co-channel) makes it possible to generate suitable artificial reference noise for other channel.
Sum up
2 of compensation method proposed by the invention and notch filter are not all:
-at first, and the frequency that the preliminary cognition in the limited ground of its needs will compensate, promptly this algorithm converges on prominent frequency automatically near initial value,
-secondly, it can pass through controlling models noise parameter σ
w 2(n) and Q (n) prevent that the Kalman filter expanded from removing the phonological component of same frequency.
The present invention proposes: realize this control by the voice activity detection (vad) method.This method is used for moving communicating field, for example referring to " Voice-Activity Detector ", (ETSI Rec.GSM 06.92,1989).Described detection method is determined threshold value.Be higher than this threshold value, that is, when going out realize voice in the signal, by as
Give bigger value to measuring noise like that, come terminal parameter to estimate.When being lower than this threshold value, that is, when no longer going out realize voice in the signal, begin parameter estimation and tracking once more.
Can also comprise information from the different sensors source, i.e. revolution meter by adding measure equation independently.In this way, even also can the tracking frequencies value in the speech process, and need not to stop to estimate.
According to the present invention, the Kalman filter of a plurality of expansions is further connected.Thus, the strongest sinusoidal perturbation of power in the given frequency range of the necessary erasure signal of first wave filter or this signal.Then the signal that is obtained is offered second wave filter that can suppress time strong sinusoidal perturbation of power.
Also propose: carry out another step and suppress the residual disturbance signal.Thus, after compensation process, can filter this signal according to the method for Ephraim and Malah.Described method is described in following document: " Speech enhancement using a minimum mean-square errorshort-time spectral amplitude estimator " (by Yariv Ephraim and David Malah, IEEE Transactions on Acoustics, Speech and Signal Processing, 32 (6), December 1984).
Claims (11)
1, a kind of method of sinusoidal perturbation of the unknown frequency that is used for eliminating the voice signal (1) that is subjected to disturbance,
May further comprise the steps:
Estimate three parameters of (2) described sinusoidal perturbation (9), described three parameters are amplitude, phase place and frequency,
Generate (4) reference signal (5) according to estimated parameter, and
From the described useful signal (1) that is subjected to disturbance, deduct (6) described reference signal (5),
It is characterized in that, Kalman filter by expansion is carried out described estimation (2), and adjusts the degree of confidence in the initial value of described estimation (2) step by the error covariance matrix that the Kalman filter of described expansion is controlled in the measurement based on speech activity.
2, method according to claim 1,
Wherein, utilize the value of additional sensor and/or learning process that initialization is carried out in the estimation (2) of the parameter of described sinusoidal perturbation (9).
3, method according to claim 1 and 2,
Wherein, the information from additional sensor is integrated, as the additional measure equation of Kalman's system.
4, method according to claim 1,
Wherein, eliminate a plurality of sinusoidal perturbations (9) by the method that repeats claim 1 continuously.
5, method according to claim 1,
Wherein, before described estimation (2) step, the described useful signal (1) of disturbance that is subjected to is carried out bandpass filtering.
6, method according to claim 5,
Wherein, before the method with claim 1 or 4 is applied to each wave band, utilize a plurality of bandpass filter that the described useful signal (1) of disturbance that is subjected to is decomposed into a plurality of frequency ranges.
7, method according to claim 6,
Wherein
Eliminate the given sinusoidal perturbation (9) in first frequency range, and
By the reference signal (5) that generates for the given sinusoidal perturbation of eliminating in described first frequency range (9), eliminate the given sinusoidal perturbation (9) in second frequency range.
8, method according to claim 7,
Wherein, by adjusting the described reference signal (5) that generates for the given sinusoidal perturbation of eliminating in described first frequency range (9), the given sinusoidal perturbation (9) in described second frequency range is eliminated the ratio of response of first band frequency and the response of second band frequency.
9, method according to claim 1,
It is characterized in that,
Described method is that time selectivity is carried out.
10, method according to claim 1,
Wherein, according to Ephraim and Malah the estimated useful signal (7) that is obtained is filtered.
11, a kind of system of sinusoidal perturbation of the unknown frequency that is used for eliminating the information carrying signal that is subjected to disturbance,
Wherein, calculation element is designed to, and realizes according to each the described method in the claim 1 to 10.
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| EP04012471 | 2004-05-26 | ||
| EP04012471.1 | 2004-05-26 | ||
| EP04024861.9 | 2004-10-19 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1702738A CN1702738A (en) | 2005-11-30 |
| CN100589184C true CN100589184C (en) | 2010-02-10 |
Family
ID=35632439
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN200510074622A Expired - Fee Related CN100589184C (en) | 2004-05-26 | 2005-05-26 | Subtractive cancellation of harmonic noise |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100589184C (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103609028A (en) * | 2011-06-16 | 2014-02-26 | 洛克希德马丁公司 | Method and system to adaptively cancel sinusoidal interference from a signal processing system |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101452698B (en) * | 2007-11-29 | 2011-06-22 | 中国科学院声学研究所 | Voice HNR automatic analytical method |
| JP5640238B2 (en) * | 2008-02-28 | 2014-12-17 | 株式会社通信放送国際研究所 | Singularity signal processing system and program thereof |
| CN103916339A (en) * | 2014-01-19 | 2014-07-09 | 中国人民解放军后勤工程学院 | ANF frequency unbiased estimation method based on gradient steepest descent |
| DE102015226365A1 (en) * | 2015-12-21 | 2017-06-22 | Robert Bosch Gmbh | Method for measuring the variance in a measuring signal, method for data fusion, computer program, machine-readable storage medium and device |
| CN106771788B (en) * | 2017-03-13 | 2019-05-21 | 信利光电股份有限公司 | A kind of processing method being disturbed sinusoidal wave data, apparatus and system |
| CN110303521B (en) * | 2018-03-27 | 2022-07-01 | 清华大学 | Robot joint torque signal acquisition system and method |
| CN109003621B (en) * | 2018-09-06 | 2021-06-04 | 广州酷狗计算机科技有限公司 | Audio processing method and device and storage medium |
-
2005
- 2005-05-26 CN CN200510074622A patent/CN100589184C/en not_active Expired - Fee Related
Non-Patent Citations (2)
| Title |
|---|
| REMOVING HUM FROM SPOKEN LANGUAGERESOURCES. HARTMUT R PF ITZINGER.PROC. OF ICSLP 2000. 2000 |
| REMOVING HUM FROM SPOKEN LANGUAGERESOURCES. HARTMUT R PF ITZINGER.PROC. OF ICSLP 2000. 2000 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103609028A (en) * | 2011-06-16 | 2014-02-26 | 洛克希德马丁公司 | Method and system to adaptively cancel sinusoidal interference from a signal processing system |
| CN103609028B (en) * | 2011-06-16 | 2016-04-06 | 洛克希德马丁公司 | In order to eliminate the method and system of sinusoidal interference from signal processing system adaptively |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1702738A (en) | 2005-11-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN100589184C (en) | Subtractive cancellation of harmonic noise | |
| US7453963B2 (en) | Subtractive cancellation of harmonic noise | |
| CN106340292B (en) | A kind of sound enhancement method based on continuing noise estimation | |
| US6289309B1 (en) | Noise spectrum tracking for speech enhancement | |
| US5806025A (en) | Method and system for adaptive filtering of speech signals using signal-to-noise ratio to choose subband filter bank | |
| EP1973104B1 (en) | Method and apparatus for estimating noise by using harmonics of a voice signal | |
| CN107408394B (en) | Determining a noise power level difference and a sound power level difference between a primary channel and a reference channel | |
| CA2310491A1 (en) | Noise suppression for low bitrate speech coder | |
| JPS62265700A (en) | Noise compensation for voice recognition | |
| US20150032445A1 (en) | Noise estimation apparatus, noise estimation method, noise estimation program, and recording medium | |
| CN103578477B (en) | Denoising method and device based on noise estimation | |
| US5963899A (en) | Method and system for region based filtering of speech | |
| Yuo et al. | Robust features for noisy speech recognition based on temporal trajectory filtering of short-time autocorrelation sequences | |
| KR101581885B1 (en) | Apparatus and Method for reducing noise in the complex spectrum | |
| DE112007003674T5 (en) | Method and apparatus for single-channel speech enhancement based on a latency-reduced auditory model | |
| WO2006114101A1 (en) | Detection of speech present in a noisy signal and speech enhancement making use thereof | |
| So et al. | A non-iterative Kalman filtering algorithm with dynamic gain adjustment for single-channel speech enhancement | |
| Dionelis et al. | Speech enhancement using modulation-domain Kalman filtering with active speech level normalized log-spectrum global priors | |
| Batina et al. | Noise power spectrum estimation for speech enhancement using an autoregressive model for speech power spectrum dynamics | |
| Rao et al. | Speech enhancement using sub-band cross-correlation compensated Wiener filter combined with harmonic regeneration | |
| Van Hamme | Robust speech recognition using cepstral domain missing data techniques and noisy masks | |
| Elshamy et al. | Two-stage speech enhancement with manipulation of the cepstral excitation | |
| Tran et al. | Speech enhancement using modified IMCRA and OMLSA methods | |
| Xie et al. | Speech enhancement by nonlinear spectral estimation-a unifying approach. | |
| Rao et al. | Speech enhancement using cross-correlation compensated multi-band wiener filter combined with harmonic regeneration |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20100210 Termination date: 20120526 |