CN107369456A - Noise cancellation method based on generalized sidelobe canceller in digital deaf-aid - Google Patents
Noise cancellation method based on generalized sidelobe canceller in digital deaf-aid Download PDFInfo
- Publication number
- CN107369456A CN107369456A CN201710540870.1A CN201710540870A CN107369456A CN 107369456 A CN107369456 A CN 107369456A CN 201710540870 A CN201710540870 A CN 201710540870A CN 107369456 A CN107369456 A CN 107369456A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msup
- msub
- signal
- lambda
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 206010011878 Deafness Diseases 0.000 title claims abstract description 18
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 37
- 238000012545 processing Methods 0.000 claims abstract description 23
- 230000001427 coherent effect Effects 0.000 claims abstract description 10
- 230000003044 adaptive effect Effects 0.000 claims abstract description 6
- 230000000903 blocking effect Effects 0.000 claims description 20
- 239000011159 matrix material Substances 0.000 claims description 20
- 230000008859 change Effects 0.000 claims description 5
- 230000017105 transposition Effects 0.000 claims description 2
- 230000006872 improvement Effects 0.000 abstract description 4
- 230000002708 enhancing effect Effects 0.000 abstract description 3
- 238000004088 simulation Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 32
- 230000000694 effects Effects 0.000 description 8
- 230000008569 process Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000007812 deficiency Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 238000012935 Averaging Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 239000004568 cement Substances 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 238000001027 hydrothermal synthesis Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000452 restraining effect Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 230000005236 sound signal Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
- G10L19/02—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using spectral analysis, e.g. transform vocoders or subband vocoders
- G10L19/0212—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using spectral analysis, e.g. transform vocoders or subband vocoders using orthogonal transformation
- G10L19/0216—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using spectral analysis, e.g. transform vocoders or subband vocoders using orthogonal transformation using wavelet decomposition
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0264—Noise filtering characterised by the type of parameter measurement, e.g. correlation techniques, zero crossing techniques or predictive techniques
-
- 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
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
- G10L2021/02161—Number of inputs available containing the signal or the noise to be suppressed
- G10L2021/02166—Microphone arrays; Beamforming
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Spectroscopy & Molecular Physics (AREA)
- General Health & Medical Sciences (AREA)
- Neurosurgery (AREA)
- Otolaryngology (AREA)
- Soundproofing, Sound Blocking, And Sound Damping (AREA)
Abstract
The invention discloses the noise cancellation method based on generalized sidelobe canceller in a kind of digital deaf-aid, the present invention is effectively combined the method for wavelet threshold denoising with beamforming algorithm, wavelet threshold denoising is first carried out to every road signal of microphone array, eliminates the coherent noise in signal with time delay error and leakage signal caused by reducing noise.In wavelet threshold denoising algorithm, threshold function table directly determines the quality of enhancing effect, propose a kind of continuous threshold function table of the improvement with adjusting parameter, the threshold function table is handled the wavelet coefficient of Noisy Speech Signal in wavelet field, the voice signal for the coherent noise that is eliminated after reconstruction processing.The adaptive beam former of generalized side lobe canceller accelerates the renewal speed of weight coefficient using improved LMS algorithm.Simulation results show that the method in the present invention effectively increases the signal to noise ratio and intelligibility of voice signal compared to traditional generalized side lobe canceller.
Description
Technical field
The present invention relates to the speech enhancement technique field of modern signal processing, based on by broad sense particularly in digital deaf-aid
The noise cancellation method of valve Canceller.
Background technology
Nowadays, requirement more and more higher of the hearing patient to digital deaf-aid, many patients complain worn audiphone without
Method is all listened clear in various environment, and it is not very comfortable to feel.Because actual listening environments are often complexity,
Sound comes from many directions, causes very big interference, therefore it is research digital hearing aid to design good audio signal processing method
The core of device.Microphone array can make full use of spatial domain, time domain and the frequency domain information of voice signal, while have high spatial point
The features such as resolution, high RST gain are with stronger antijamming capability.Due to surrounding environment existing interference in itself, along with electronics
Influence caused by when device works, the signal for causing microphone to receive all is to mix noisy mixed signal, therefore use has
The method of effect removes the ambient noise in voice to improve signal to noise ratio as the primary study object in Speech processing.Tradition
Voice enhancement algorithm have spectrum-subtraction, wavelet threshold denoising algorithm, adaptive null method and signal subspace method etc..
Griffiths etc. proposes generalized sidelobe canceller (Generalized sidelobe canceller, GSC), broad sense first
The beamforming algorithm of Sidelobe Canceller structure has high-performance and relatively low amount of calculation, extensively should be able in digital deaf-aid
With.GSC structures include the fixed beam former (Fixed Beamformer, FBF) of upper branch road, the blocking matrix of lower branch road
(Blocking Matrix, BM) and adaptive-filtering, and multi input Canceller (Multiple-input Canceller,
MC), for eliminating the residual noise of fixed beam former output end.But fixed beam formation algorithm is laggard to delay compensation
Row weighted sum, which is averaged, can eliminate incoherent noise, but mixed coherent noise can influence time delay and estimate to obtain accuracy；Cause
This Xian Duimei roads signal, which carries out wavelet threshold denoising, can eliminate coherent noise enhancing accuracy.
Wavelet threshold denoising method includes two classical functions of hard -threshold and soft-threshold, but hard threshold function can go out at threshold value
Existing discontinuous shortcoming is so can influence the result of de-noising in reconstruction signal.Soft-threshold function wavelet field wavelet coefficient and
Original wavelet coefficient has certain error so can influence the intensity of signal in reconstruction signal.
The content of the invention
The technical problems to be solved by the invention are overcome the deficiencies in the prior art and provided in digital deaf-aid based on wide
The noise cancellation method of adopted Sidelobe Canceller, coherent noise can also be eliminated by eliminating incoherent noise well, make digital hearing aid
Signal obtained by device has more preferable speech understanding degree.
The present invention uses following technical scheme to solve above-mentioned technical problem：
According to the noise cancellation method based on generalized sidelobe canceller in digital deaf-aid proposed by the present invention, including it is following
Step：
Step 1, wavelet threshold processing is carried out to the Noisy Speech Signal that microphone array receives；Wavelet threshold processing is to adopt
Processing is reconstructed with the wavelet coefficient of improved wavelet threshold function pair Noisy Speech Signal, so as to eliminate Noisy Speech Signal
In coherent noise；
The improved wavelet threshold function is
Wherein, Wj,kThe wavelet coefficient of progress jth k-th of signals with noise of layer in wavelet field is represented,It is according to small echo
The wavelet coefficient of voice signal after threshold function table processing,It is the threshold value of threshold function table, σ is being averaged for noise
Variance, N are the length for the Noisy Speech Signal that microphone array is received, and e is the nature truth of a matter, and a is adjusting parameter；
Step 2, delay compensation is carried out to the voice signal after wavelet threshold processing, the voice signal after delay compensation is entered
Row processing, produces reference speech signal yF(t)；
yF(t)=WTX(t)
Wherein, W=[w1,w2,…,wM]TIt is fixed weight vector, wqIt is the fixed power of microphone array q roads reception signal
Coefficient, 1≤q≤M, M are that microphone array shakes first total number, and subscript T is transposition, X (t)=[x1'(t),x2'(t),…,xM'
(t)]TIt is the array received signal vector after delay compensation, xq' (t) be q roads voice signal after delay compensation；
The desired signal in the voice signal after delay compensation is filtered out using blocking matrix B, produces noise reference signal
yB(t)；
Step 3, obtain former targeted voice signal y (t), y (t)=yF(t)-yB(t)。
It is further excellent as the noise cancellation method based on generalized sidelobe canceller in digital deaf-aid of the present invention
Change scheme,
yB(t)=AT(k) U (t), A (k) are sef-adapting filter vector A k-th of coefficient, after blocking matrix is handled
Array signal U (t)=BX (t), wherein, blocking matrix
It is further excellent as the noise cancellation method based on generalized sidelobe canceller in digital deaf-aid of the present invention
Change scheme, A=[a1,a2,…,aM]T, aqIt is q-th of coefficient of sef-adapting filter.
It is further excellent as the noise cancellation method based on generalized sidelobe canceller in digital deaf-aid of the present invention
Change scheme, adaptive filter coefficient vector A optimal solution is found using improved LMS algorithm, make noise output power most
It is small；Relation using+1 coefficient of improved LMS algorithm sef-adapting filter kth and k-th of coefficient is as follows：
It is further excellent as the noise cancellation method based on generalized sidelobe canceller in digital deaf-aid of the present invention
Change scheme, acquisition a optimum value is adjusted according to least mean-square error MSE.
The present invention compared with prior art, has following technique effect using above technical scheme：
(1) present invention carries out wavelet threshold denoising by Xian Duimei roads signal makes time delay estimation more accurate, closely passes through
Delay compensation Hou Ge roads signal can be very good to align, and can eliminate coherent noise well, compensate for traditional delay-ask
With the deficiency of Beamforming Method, it is eliminated incoherent noise or eliminate coherent noise, increase its practicality；
(2) improve LMS algorithm and can also quickly update weight coefficient and carry out real-time tracing signal and signals with noise is filtered,
Accelerate to make steady-state error tend towards stability while convergence, misalignment rate is effectively controlled, so that the signal obtained by digital deaf-aid
With more preferable speech understanding degree.
Brief description of the drawings
Fig. 1 improves GSC structured flowcharts for the present invention.
Fig. 2 is Wj,k>When 0, comparison that is soft, hard and improving threshold function table.
Fig. 3 is different wavelet threshold function process signal waveforms；Wherein, (a) is dye noise cancellation signal, and (b) is represented at hard -threshold
Signal after reason, (c) represent the signal of soft-threshold processing, and (d) represents the signal proposed by the present invention for improving threshold function table processing.
Fig. 4 is that each algorithm output signal compares；Wherein, (a) is noisy signal, and (b) is the output letter of traditional GSC structures
Number, (c) is the output signal after generalized sidelobe cancellation algorithms are improved to blocking matrix, and (d) is innovatory algorithm of the present invention
The output signal of GSC structures.
Embodiment
Technical scheme is described in further detail below in conjunction with the accompanying drawings：
Traditional GSC structures mainly include three parts, and Part I is fixed beam former (Fixed
Beamformer, FBF), for producing speech reference signal；The second part be blocking matrix (Blocking Matrix,
BM), for producing noise reference signal；Part III is multi input Canceller (Multiple-input Canceller, MC),
For eliminating the residual noise of fixed beam former output end.
Using the inventive method specifically according to following steps：
Step 1, the every road signal received to microphone array are improved wavelet threshold function Wavelet Denoising Method and handled
To the voice signal for removing coherent noise：
X (t)=[x in Fig. 11(t),x2(t),…,xM(t)]TFor array signal vector, the main of wavelet threshold denoising is appointed
Business is that the wavelet coefficient with voice is remained with wavelet transformed domain, the wavelet coefficient of cancelling noise.Because signals with noise is small
After Wave Decomposition, in wavelet field, coefficient corresponding to useful signal is very big, and coefficient very little corresponding to noise, and noise is in wavelet field pair
The coefficient answered still meets that Gauss white noise is distributed, so appropriate threshold value is chosen according to certain rule, to the coefficient less than threshold value
Zero is gradually reduced to, so as to the wavelet coefficient of cancelling noise.Wavelet coefficient after threshold process is reconstructed, it is possible to obtain
Signal after denoising.There is relative time delay τ between the signal that each microphone receivesi, it is necessary to delay compensation is carried out to each letter
Number plus the timeTo offset relative time delay,It can be tried to achieve by delay time estimation method, wavelet threshold denoising is mutually tied with it
Close, Wavelet Denoising Method first is carried out to each channel speech signal before time delay estimation, not only can first eliminate partial noise, and
Increase the accuracy of time delay estimation, compensation Hou Ge roads signal is preferably alignd.Preposition wavelet threshold denoising it is general
Step is：
1) select wavelet function to pretreatment Noisy Speech Signal progress wavelet transformation.
2) threshold value λ, and select threshold function table to handle original wavelet coefficients.
3) it is wavelet reconstruction to carry out inverse transformation to the wavelet coefficient after processing.
4) signal after wavelet threshold denoising is sent into BM modules and MC modules obtains GSC pattern handlings and obtain output letter
Number.
From above step it can be seen that in wavelet threshold denoising, the selection of threshold function table has direct shadow to wavelet coefficient
Ring so as to have impact on the effect of de-noising, the threshold function table of good properties can reject more noise signals, adapt to different noises
The noisy speech of ratio.Improved wavelet threshold function is
Wherein Wj,kIt is the wavelet coefficient of signals with noise,It is according to the wavelet coefficient after threshold function table and threshold process.σ is the average variance of noise, and N is signal length.Wherein a is adjusting parameter.Function is improved when a tends to be infinite
Tend to hard threshold function, adjusting parameter improves the flexibility of threshold function table.But also generate a should be specifically configured to it is how many
Problem.What the present invention took is AFSA methods, and this method can adjust acquisition a optimum value according to least mean-square error MSE.
In order to verify the function it is available under, we do analysis below to the continuity of a function：
Work as Wj,k>During λ,
When | Wj,kDuring |≤λ,
Work as Wj,kDuring ＜-λ,
SoI.e. in Wj,kDuring=λ,It is continuous, institute
WithAll it is continuous in whole domain.
Hard threshold function discontinuously causes the generation for shaking noise in Fig. 2, and soft-threshold function is to be less than threshold in wavelet coefficient
Direct zero setting during value, although continuously introducing fixed error, threshold function table proposed by the present invention connects in whole domain
It is continuous, smooth processing has been carried out to wavelet coefficient, the part of some weak voices has been remained in wavelet field, avoids droop
Produce, improve the quality of voice after reconstruct.
To the oscillogram we show both dye noise cancellation signal after the processing of different threshold function tables in Fig. 3.Abscissa table in Fig. 3
Show sampling number/103, ordinate expression signal amplitude/mV.(a) in Fig. 3 represents dye noise cancellation signal, and (b) in Fig. 3 represents hard
Signal after threshold process, because the inherent defect of hard threshold function result in the Pseudo-Gibbs artifacts of reconfiguration waveform figure, in Fig. 3
(c) represent the signal of soft-threshold processing, signal is more smooth but can lack some non-stationary characteristics, and (d) in Fig. 3 is represented
Improvement threshold function table proposed by the present invention, the shortcomings that compensate for soft or hard function, preferably handle signals with noise.
By Fig. 2 and Fig. 3 analysis, improving wavelet threshold function can be in small echo compared with soft-threshold and hard -threshold
The processing preferably to weak voice wavelet coefficient and noise separation is realized on domain, reducing fixed error realizes reconstruction signal
Smoothing processing, there is preferable noise processed effect.
Step 2, pass through lower branch road BM modules acquisition noise reference signal：
The effect of blocking matrix B in Fig. 1 is to filter out the desired signal in reception signal, produces noise reference signal.
The signal that microphone receives have passed through time delay estimation, then make the expectation in the signal of each microphone reception by delay compensation
Signal is completely in phase, so only leave noise signal to block to fall desired signal, the element in every a line in blocking matrix
Sum is necessary for zero, output below is free of desired signal.IfFor the i-th row element in blocking matrix B to
Amount.As long as then meet for all i：
Because each bmVector is mutual Line independent, and it is individual linear only that the signal after blocking matrix is up to (M-1)
Vertical element, thus, blocking matrix B row vector number must be (M-1) or less.One of them meets above-mentioned condition
Conventional blocking matrix is：
Signal after blocking matrix is handled is：
U=BX (5)
Formula (5) can, which is found out, does not contain desired signal in signal vector, comprise only noise signal.If sef-adapting filter
Coefficient vector be A=[a1,a2,…,aM]T, its export noise reference signal be：
Step 3, the output signal of MC modules namely it is expected that speech reference signal subtracts the difference obtained by noise reference signal：
Y (t)=yF(t)-yB(t) (7)
Due to yF(t) desired signal composition and part interfering noise signal composition are included in, and yB(t) interference is comprised only in
Noise component, just it is expected voice signal containing only original by subtracting each other y (t).In order that noise output power is minimum, it is necessary to find certainly
The optimal solution of adaptive filter coefficient vector, adjusted using the weight vector of improved LMS algorithm by formula (10) and (11)：
A ' (the k)=μ y of A (k)+2 (t) U (t) (8)
A (the k+1)=μ y ' of A ' (k)+2 (t) U (t) (9)
Wherein, μ is the arithmetic number that step factor is control algolithm convergence rate and stability, in formula
Y ' (t)=yF(t)-yB′(t),y′B(t)=[A ' (k)]TU(t) (10)
Compare (8) and (9) and understand that the A ' (k) for improving LMS algorithm accelerates power equivalent to A (k+1) of traditional LMS algorithm
The iteration speed of vector, (8)-(10) substitution (11) formula is obtained：
A(k+1)
=μ the y of A (k)+2 (t) U (t)+2 μ { yF(t)-[A(k)+2μy(t)U(t)]TU(t)}U(t)
=μ the y of A (k)+2 (t) U (t) -2 μ { yF(t)-yB(t)-2μy(t)||U(t)||2}U(t)
=A (k)+4 μ [1- μ | | U (t) | |2]y(t)U(t) (11)
Compared to traditional LMS algorithm, it is believed that the μ in this algorithm1=2 μ [1- μ | | U (t) | |2] original μ is instead of,
μ1Maximum beIn order to take into account convergence rate and imbalance then μ=μ1, finally improve the iteration calculation of the power system of algorithm
Method such as (12) formula
Compared with the LMS algorithm of fixed step size, improvement LMS algorithm of the invention can quickly update weight coefficient and chase after in real time
Track signal is filtered to signals with noise, steady-state error is tended towards stability while accelerating and restraining, is effectively controlled misalignment rate.
Convergence rate and steady output rate are taken into account, so improved LMS algorithm is relatively specific for handling nonstationary random signal.
The one section of pure voice signal selected in experiment in sound bank, sample frequency are 16kHz. respectively with traditional GSC
Output signal emulation experiment such as Fig. 4 is obtained after structure algorithm, method and improved structure of the present invention processing：
In Fig. 4, abscissa represents sampling number/104, ordinate expression signal amplitude/mV.(a) in Fig. 4 is to add letter of making an uproar
Number, (b) in Fig. 4 is that (c) in the output and Fig. 4 of traditional GSC structures is output signal after being improved to blocking matrix,
(d) in Fig. 4 is that the output of the GSC structures of innovatory algorithm of the present invention compares it can be found that the improved GSC structures pair of the present invention
The voice leakage problem of traditional GSC structures has certain compensation, and certain enhancing effect has been seen on waveform.Have from objective side
In face of the signal to noise ratio and mean square error than output signal, the SNR after de-noising is more big, illustrates to make an uproar relative to the intensity of useful signal
The interference of sound is smaller, then the effect of de-noising is better.The enhanced signal of the smaller explanation of mean square error MSE values after de-noising and original
Error between signal is smaller, and de-noising effect is better.Result of calculation such as table 1, the results showed that compared with traditional GSC algorithms, this hair
The signal to noise ratio of bright algorithm is higher, and mean square error is smaller.The effect of de-noising is relatively preferable.
The signal to noise ratio and mean square error of the output signal of table 1
| SNR/dB | MSE | |
| Original signal to noise ratio | 10.9329 | 1.2542 |
| Traditional GSC algorithms | 15.7186 | 1.1480 |
| Improve blocking matrix algorithm | 15.8588 | 1.1296 |
| Inventive algorithm | 16.0466 | 1.1055 |
Using Mean Opinion Score MOS (Mean Opinion Scores) method come to output signal for subjective aspect
Definition, comfort level and intelligibility evaluate.Tied in the environment of being respectively 0,3,5 in signal to noise ratio by 150 words by GSC
After structure, battery of tests person is allowed to recognize；Similarly, this 150 words are passed through into innovatory algorithm under same state of signal-to-noise respectively
After GSC structures, another group of experimenter is allowed to recognize.2. results are shown in Table as a result to every group of scoring averaging to illustrate different
New make is above traditional GSC algorithms than lower inventive algorithm, illustrates that modified hydrothermal process improves the identification of voice, improves
Voice quality.
The MOS scorings of the output signal of table 2
The present invention have studied the speech enhan-cement that the wavelet threshold denoising based on GSC structures and improvement threshold function table is combined and calculate
Method, situation and the tradition that can influence time delay estimation and increase voice leakage before GSC pattern handlings for noisy signal are adaptive
Improved wavelet threshold function and improved LMS algorithm are proposed in terms of the renewal speed for answering algorithm.The simulation experiment result shows
The algorithm realize reduce GSC structure output signals residual noise and voice leakage situation, improve signal signal to noise ratio and
Voice quality.
Above content is to combine specific preferred embodiment further description made for the present invention, it is impossible to is assert
The specific implementation of the present invention is confined to these explanations.For general technical staff of the technical field of the invention,
On the premise of not departing from present inventive concept, some simple deductions can also be made or substituted, should all be considered as belonging to the present invention's
Protection domain.
Claims (5)
1. the noise cancellation method based on generalized sidelobe canceller in digital deaf-aid, it is characterised in that comprise the following steps：
Step 1, wavelet threshold processing is carried out to the Noisy Speech Signal that microphone array receives；Wavelet threshold processing is to use to change
Processing is reconstructed in the wavelet coefficient of the wavelet threshold function pair Noisy Speech Signal entered, so as to eliminate in Noisy Speech Signal
Coherent noise；
The improved wavelet threshold function is
<mrow>
<msub>
<mover>
<mi>W</mi>
<mo>~</mo>
</mover>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>=</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>-</mo>
<mi>&lambda;</mi>
<mo>+</mo>
<mi>&lambda;</mi>
<mrow>
<mo>(</mo>
<mfrac>
<mrow>
<msup>
<mi>e</mi>
<mrow>
<msup>
<mi>a</mi>
<mn>2</mn>
</msup>
<mrow>
<mo>(</mo>
<mrow>
<mrow>
<mo>(</mo>
<mrow>
<msub>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>-</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
<mo>/</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
</msup>
<mo>-</mo>
<msup>
<mi>e</mi>
<mrow>
<mo>-</mo>
<msup>
<mi>a</mi>
<mn>2</mn>
</msup>
<mrow>
<mo>(</mo>
<mrow>
<mrow>
<mo>(</mo>
<mrow>
<msub>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>-</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
<mo>/</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
</msup>
</mrow>
<mrow>
<msup>
<mi>e</mi>
<mrow>
<msup>
<mi>a</mi>
<mn>2</mn>
</msup>
<mrow>
<mo>(</mo>
<mrow>
<mrow>
<mo>(</mo>
<mrow>
<msub>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>-</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
<mo>/</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
</msup>
<mo>+</mo>
<msup>
<mi>e</mi>
<mrow>
<mo>-</mo>
<msup>
<mi>a</mi>
<mn>2</mn>
</msup>
<mrow>
<mo>(</mo>
<mrow>
<mrow>
<mo>(</mo>
<mrow>
<msub>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>-</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
<mo>/</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
</msup>
</mrow>
</mfrac>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
<mtd>
<mrow>
<msub>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>></mo>
<mi>&lambda;</mi>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>0</mn>
</mtd>
<mtd>
<mrow>
<mo>|</mo>
<msub>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>|</mo>
<mo>&le;</mo>
<mi>&lambda;</mi>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>+</mo>
<mi>&lambda;</mi>
<mo>+</mo>
<mi>&lambda;</mi>
<mrow>
<mo>(</mo>
<mfrac>
<mrow>
<msup>
<mi>e</mi>
<mrow>
<msup>
<mi>a</mi>
<mn>2</mn>
</msup>
<mrow>
<mo>(</mo>
<mrow>
<mrow>
<mo>(</mo>
<mrow>
<msub>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>+</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
<mo>/</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
</msup>
<mo>-</mo>
<msup>
<mi>e</mi>
<mrow>
<mo>-</mo>
<msup>
<mi>a</mi>
<mn>2</mn>
</msup>
<mrow>
<mo>(</mo>
<mrow>
<mrow>
<mo>(</mo>
<mrow>
<msub>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>+</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
<mo>/</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
</msup>
</mrow>
<mrow>
<msup>
<mi>e</mi>
<mrow>
<msup>
<mi>a</mi>
<mn>2</mn>
</msup>
<mrow>
<mo>(</mo>
<mrow>
<mrow>
<mo>(</mo>
<mrow>
<msub>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>+</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
<mo>/</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
</msup>
<mo>+</mo>
<msup>
<mi>e</mi>
<mrow>
<mo>-</mo>
<msup>
<mi>a</mi>
<mn>2</mn>
</msup>
<mrow>
<mo>(</mo>
<mrow>
<mrow>
<mo>(</mo>
<mrow>
<msub>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo>+</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
<mo>/</mo>
<mi>&lambda;</mi>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
</msup>
</mrow>
</mfrac>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
<mtd>
<mrow>
<msub>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
</msub>
<mo><</mo>
<mo>-</mo>
<mi>&lambda;</mi>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
Wherein, Wj,kThe wavelet coefficient of progress jth k-th of signals with noise of layer in wavelet field is represented,It is according to wavelet threshold
The wavelet coefficient of voice signal after function processing,It is the threshold value of threshold function table, σ is the average variance of noise,
N is the length for the Noisy Speech Signal that microphone array is received, and e is the nature truth of a matter, and a is adjusting parameter；
Step 2, delay compensation is carried out to the voice signal after wavelet threshold processing, at the voice signal after delay compensation
Reason, produce reference speech signal yF(t)；
yF(t)=WTX(t)
Wherein, W=[w1,w2,…,wM]TIt is fixed weight vector, wqIt is the fixation weight coefficient of microphone array q roads reception signal,
1≤q≤M, M are that microphone array shakes first total number, and subscript T is transposition, X (t)=[x1'(t),x2'(t),…,xM'(t)]TIt is
Array received signal vector after delay compensation, xq' (t) be q roads voice signal after delay compensation；
The desired signal in the voice signal after delay compensation is filtered out using blocking matrix B, produces noise reference signal yB(t)；
Step 3, obtain former targeted voice signal y (t), y (t)=yF(t)-yB(t)。
2. the noise cancellation method based on generalized sidelobe canceller in digital deaf-aid according to claim 1, its feature
It is,
yB(t)=AT(k) U (t), A (k) is sef-adapting filter vector A k-th of coefficients, the array after blocking matrix is handled
Signal U (t)=BX (t), wherein, blocking matrix
3. the noise cancellation method based on generalized sidelobe canceller in digital deaf-aid according to claim 2, its feature
It is, A=[a1,a2,…,aM]T, aqIt is q-th of coefficient of sef-adapting filter.
4. the noise cancellation method based on generalized sidelobe canceller in digital deaf-aid according to claim 2, its feature
It is, adaptive filter coefficient vector A optimal solution is found using improved LMS algorithm, makes noise output power minimum；
Relation using+1 coefficient of improved LMS algorithm sef-adapting filter kth and k-th of coefficient is as follows：
<mrow>
<mi>A</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>A</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mfrac>
<mrow>
<mi>y</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mo>|</mo>
<mo>|</mo>
<mi>U</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>|</mo>
<msup>
<mo>|</mo>
<mn>2</mn>
</msup>
</mrow>
</mfrac>
<mi>U</mi>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
<mo>.</mo>
</mrow>
5. the noise cancellation method based on generalized sidelobe canceller in digital deaf-aid according to claim 1, its feature
It is, acquisition a optimum value is adjusted according to least mean-square error MSE.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710540870.1A CN107369456A (en) | 2017-07-05 | 2017-07-05 | Noise cancellation method based on generalized sidelobe canceller in digital deaf-aid |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710540870.1A CN107369456A (en) | 2017-07-05 | 2017-07-05 | Noise cancellation method based on generalized sidelobe canceller in digital deaf-aid |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107369456A true CN107369456A (en) | 2017-11-21 |
Family
ID=60305915
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710540870.1A Pending CN107369456A (en) | 2017-07-05 | 2017-07-05 | Noise cancellation method based on generalized sidelobe canceller in digital deaf-aid |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107369456A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108957392A (en) * | 2018-04-16 | 2018-12-07 | 深圳市沃特沃德股份有限公司 | Sounnd source direction estimation method and device |
| CN113470681A (en) * | 2021-05-21 | 2021-10-01 | 中科上声（苏州）电子有限公司 | Pickup method of microphone array, electronic equipment and storage medium |
| CN113711303A (en) * | 2018-11-29 | 2021-11-26 | Ams传感器英国有限公司 | Method for tuning an audio system supporting noise cancellation and audio system supporting noise cancellation |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102664023A (en) * | 2012-04-26 | 2012-09-12 | 南京邮电大学 | Method for optimizing speech enhancement of microphone array |
-
2017
- 2017-07-05 CN CN201710540870.1A patent/CN107369456A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102664023A (en) * | 2012-04-26 | 2012-09-12 | 南京邮电大学 | Method for optimizing speech enhancement of microphone array |
Non-Patent Citations (6)
| Title |
|---|
| HE CAN等: ""A New Wavelet Thresholding Function Based on Hyperbolic Tangent Function"", 《MATHEMATICAL PROBLEMS IN ENGINEERING》 * |
| 刘凤山等: ""改进小波阈值函数的语音增强算法研究"", 《信号处理》 * |
| 徐进: ""一种新的麦克风阵列自适应语音增强方法"", 《应用科学学报》 * |
| 沐城: ""数字助听器中语音增强及响度补偿算法的研究"", 《中国优秀硕士学位论文全文数据库医药卫生科技辑》 * |
| 王辉等: ""数字助听器中广义旁瓣抵消器结构的汉语语音处理技术"", 《声学学报》 * |
| 高婉贞: ""数字助听器中基于麦克风阵列语音增强技术的研究"", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108957392A (en) * | 2018-04-16 | 2018-12-07 | 深圳市沃特沃德股份有限公司 | Sounnd source direction estimation method and device |
| CN113711303A (en) * | 2018-11-29 | 2021-11-26 | Ams传感器英国有限公司 | Method for tuning an audio system supporting noise cancellation and audio system supporting noise cancellation |
| CN113711303B (en) * | 2018-11-29 | 2024-02-20 | Ams传感器英国有限公司 | Method for tuning an audio system supporting noise cancellation and audio system supporting noise cancellation |
| CN113470681A (en) * | 2021-05-21 | 2021-10-01 | 中科上声（苏州）电子有限公司 | Pickup method of microphone array, electronic equipment and storage medium |
| CN113470681B (en) * | 2021-05-21 | 2023-09-29 | 中科上声（苏州）电子有限公司 | Pickup method of microphone array, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| DE102017102134B4 (en) | Globally optimized post-filtering using the least squares method for speech enhancement | |
| CN108831495B (en) | Speech enhancement method applied to speech recognition in noise environment | |
| CN106782590A (en) | Based on microphone array Beamforming Method under reverberant ambiance | |
| CN101685638B (en) | Method and device for enhancing voice signals | |
| CN105869651B (en) | Binary channels Wave beam forming sound enhancement method based on noise mixing coherence | |
| CN110148420A (en) | A kind of audio recognition method suitable under noise circumstance | |
| Ren et al. | A Causal U-Net Based Neural Beamforming Network for Real-Time Multi-Channel Speech Enhancement. | |
| CN107369456A (en) | Noise cancellation method based on generalized sidelobe canceller in digital deaf-aid | |
| CN104835503A (en) | Improved GSC self-adaptive speech enhancement method | |
| CN114694670A (en) | Multi-task network-based microphone array speech enhancement system and method | |
| CN111816200B (en) | Multi-channel speech enhancement method based on time-frequency domain binary mask | |
| CN106161820B (en) | A kind of interchannel decorrelation method for stereo acoustic echo canceler | |
| Mahmoudi et al. | Combined Wiener and coherence filtering in wavelet domain for microphone array speech enhancement | |
| Kim et al. | Factorized MVDR deep beamforming for multi-channel speech enhancement | |
| CN103971697B (en) | Sound enhancement method based on non-local mean filtering | |
| Jabloun et al. | A multi-microphone signal subspace approach for speech enhancement | |
| Xu et al. | Adaptive speech enhancement algorithm based on first-order differential microphone array | |
| CN113409810B (en) | Echo cancellation method for joint dereverberation | |
| Yu et al. | Speech enhancement based on the generalized sidelobe cancellation and spectral subtraction for a microphone array | |
| Yu | Post-filter optimization for multichannel automotive speech enhancement | |
| Zohourian et al. | Multi-channel speaker localization and separation using a model-based GSC and an inertial measurement unit | |
| CN114724574A (en) | Double-microphone noise reduction method with adjustable expected sound source direction | |
| Özen et al. | Speech noise reduction with wavelet transform domain adaptive filters | |
| Jinzai et al. | Wavelength proportional arrangement of virtual microphones based on interpolation/extrapolation for underdetermined speech enhancement | |
| Ortega-Garcia et al. | Providing single and multi-channel acoustical robustness to speaker identification systems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171121 |