CN105679330B - Based on the digital deaf-aid noise-reduction method for improving subband signal-to-noise ratio (SNR) estimation - Google Patents
Based on the digital deaf-aid noise-reduction method for improving subband signal-to-noise ratio (SNR) estimation Download PDFInfo
- Publication number
- CN105679330B CN105679330B CN201610150663.0A CN201610150663A CN105679330B CN 105679330 B CN105679330 B CN 105679330B CN 201610150663 A CN201610150663 A CN 201610150663A CN 105679330 B CN105679330 B CN 105679330B
- Authority
- CN
- China
- Prior art keywords
- subband
- signal
- noise
- frame
- snr
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 64
- 206010011878 Deafness Diseases 0.000 title claims abstract description 21
- 238000005314 correlation function Methods 0.000 claims abstract description 11
- 230000006872 improvement Effects 0.000 claims abstract description 10
- 230000015572 biosynthetic process Effects 0.000 claims abstract description 6
- 238000003786 synthesis reaction Methods 0.000 claims abstract description 6
- 238000011946 reduction process Methods 0.000 claims abstract description 4
- 230000006870 function Effects 0.000 claims description 17
- 238000001228 spectrum Methods 0.000 claims description 14
- 239000011159 matrix material Substances 0.000 claims description 11
- 238000005070 sampling Methods 0.000 claims description 11
- 238000004458 analytical method Methods 0.000 claims description 7
- 238000011156 evaluation Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 230000002194 synthesizing effect Effects 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 2
- 238000009432 framing Methods 0.000 claims 1
- 230000013707 sensory perception of sound Effects 0.000 abstract description 16
- 230000002401 inhibitory effect Effects 0.000 abstract description 2
- 238000004422 calculation algorithm Methods 0.000 description 11
- 230000009467 reduction Effects 0.000 description 8
- 239000004568 cement Substances 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 210000000613 ear canal Anatomy 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000000926 separation method Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 230000005236 sound signal Effects 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 208000027418 Wounds and injury Diseases 0.000 description 1
- 230000002238 attenuated effect Effects 0.000 description 1
- 230000004888 barrier function Effects 0.000 description 1
- 210000003477 cochlea Anatomy 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000010370 hearing loss Effects 0.000 description 1
- 231100000888 hearing loss Toxicity 0.000 description 1
- 208000016354 hearing loss disease Diseases 0.000 description 1
- 208000014674 injury Diseases 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000000873 masking effect Effects 0.000 description 1
- 238000011056 performance test Methods 0.000 description 1
- 238000013441 quality evaluation Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000003313 weakening effect Effects 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
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
-
- 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
- H04R25/505—Customised settings for obtaining desired overall acoustical characteristics using digital signal processing
Landscapes
- Engineering & Computer Science (AREA)
- Acoustics & Sound (AREA)
- Physics & Mathematics (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Quality & Reliability (AREA)
- Computational Linguistics (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Neurosurgery (AREA)
- Otolaryngology (AREA)
- Noise Elimination (AREA)
Abstract
The present invention provides a kind of digital deaf-aid noise-reduction method based on improvement subband signal-to-noise ratio (SNR) estimation, and original signal is resolved into several subbands using resolution filter group；Then the cross-correlation function of adjacent two frame signal and respective mean-square value in each subband are calculated, to estimate the signal-to-noise ratio of the subband；Secondly it according to each subband gain of signal-to-noise ratio computation estimated, and is multiplied to obtain modified subband signal with subband signal；Modified each subband signal synthesis is finally obtained to the voice after noise reduction process.More accurately subband signal-noise ratio estimation method makes ambient noise inhibitory effect more preferable to this method, mitigates the auditory fatigue of hearing aid user；Method is simple and efficient, and avoids inverse Fourier transform and time delay performance is largely improved, and more traditional spectrum-subtraction improves 60.6%, and more traditional Wiener Filter Method improves 40.7%.
Description
Technical field
The present invention relates to a kind of based on the digital deaf-aid noise-reduction method for improving subband signal-to-noise ratio (SNR) estimation.
Background technique
Hearing loss is to influence the serious disease of human lives.Long-term dysaudia can also in addition to influencing daily exchange
Cause psychological problems, brings heavy burden to society and family.Wear hearing aid is that light-severe listens that barrier patient's is most effective
The means of hearing intervention and rehabilitation.Understand that voice difficulty is to perplex the most common problem of hearing aid user in noise circumstance.
In order to improve speech understanding degree and listen to comfort level, hearing aid speech enhancement technique can substantially be attributed to two classes: shotgun microphone
And noise-reduction method.The former is the otherness design based on voice and noise spatially, utilization orientation microphone or wave beam shape
Carry out the voice signal on Enhancement feature direction at technology.But such method is limited by the quantity or size of microphone, performance changes
It is apt to limited, and is not suitable for deep ear canal hearing aid.Second method is intended to using on the time and frequency spectrum of voice and noise
Difference separates voice from signals and associated noises.But voice and noise may have overlapping on time and frequency spectrum, because
This many researcher has made intensive studies for this problem.
In terms of speech enhan-cement, researcher proposes the significant method of many noise reduction effects.Patent US 2012/8204263
B2“Method of estimating weighting function of audio signals in a hearing aid”
I.e. a kind of hearing aid audio signal Enhancement Method realizes speech enhan-cement effect using multi-microphone Collect jointly ambient sound, but
To wearing in ear canal or for the patient of neighbouring small-sized custom hearing aid, this kind carries out speech enhan-cement but by multi-microphone
It is not applicable；Patent US 2014/13198739.8, " Enhanced dynamics processing of streaming
Audio by source separation and remixing " is i.e. a kind of to use source separation and mixed enhancing audio stream side
Method, using Non-negative Matrix Factorization theory, by carrying out study voice and noise model to clean speech data and noise data, but
This method complexity and to the more demanding of equipment；200510086877.8 " the speech enhan-cement side for hearing aid patent CN
Method " is adjusted filtering parameter using auditory masking curve, but this method need to human auditory system physiological models recognize into
One step understands, and this method usually contains the operation of some complexity such as a large amount of exponent arithmetic, logarithm operation.Above-mentioned these
Although method has certain speech enhan-cement effect, but be not particularly suited for in real-time and the very high hearing aid of power consumption requirements.
Summary of the invention
The object of the present invention is to provide a kind of based on the digital deaf-aid noise-reduction method for improving subband signal-to-noise ratio (SNR) estimation, solves
Hearing aid speech enhancement technique is existing in the prior art or the quantity by microphone or size are limited, and performance improvement is limited, and
And deep ear canal hearing aid or voice are not suitable for it and noise may have overlapping on time and frequency spectrum.
In order to achieve the above objectives, the present invention adopts the following technical scheme that:
A kind of digital deaf-aid noise-reduction method based on improvement subband signal-to-noise ratio (SNR) estimation, comprising:
Original signal is resolved by several subbands using resolution filter group, then calculates adjacent two frame in each subband
The cross-correlation function of signal and respective mean-square value, to estimate the signal-to-noise ratio of the subband；
Secondly, according to each subband gain of the signal-to-noise ratio computation estimated, and be multiplied to obtain modified subband with subband signal
Signal；
Finally, revised each subband signal synthesis to be obtained to the voice after noise reduction process.
Further, specifically includes the following steps:
S1, the microphone input signal x (n) of digital deaf-aid is AD converted, and the digital signal after conversion is divided
Frame xi(n) (i=0,1,2 ... N-1)；
S2, pass through resolution filter: every frame signal is passed through to 6 rank IIR resolution filter group h of multichanneli(n) (i=0,
1 ... M-1), M is port number, passband ripple 0.5dB, with ciIt is down-sampled for the progress of down-sampled coefficient, obtain M subband signal: x
(i, m) (i=0,1,2 ... N-1) (m=0,1,2 ... M-1)；
S3, subband signal-to-noise ratio (SNR) estimation: the signal-to-noise ratio (SNR) estimation value of m-th of the i-th frame of subband is
In formula, P (i, m), P (i-1, m) are the pure voice signal power of the i-th and (i-1)-th frame in m-th of subband, σ2(m) it is
Noise variance in m-th of subband, α are smoothing factor, 0 < α < 1.
S4, yield value calculate: bringing the Signal to Noise Ratio (SNR) (i, m) estimated into gain function gdBIn (i, m), and by gdB
(i, m) transforms to amplitude domain g (i, m).
S5, subband noise inhibit: carrying out noise attentuation to primary speech signal by the yield value of obtained each subband, obtain
To enhanced voice signal: yim(n)=g (i, m) x (i, m)；
The comprehensive output of S6, subband signal: by the output signal y in M channelim(n) (i=0,1 ... M-1) first with ciTo rise
Downsampling factor carries out a liter sampling, is finally synthesizing the voice signal y (n) exported.
Further, in step S3, subband signal-to-noise ratio (SNR) estimation specifically:
S31, the i-th and (i-1)-th frame signal defined on kth Frequency point are adjacent signals vector Xadjoin(i, k) calculates phase
The autocorrelation matrix R (i, k) of adjacent signal phasor；
S32, E [X in correlation matrix R (i, m) is estimated using L frequency spectrum in each subband2(i,m)]、E[X2
(i-1, m)] and E [X (i-1, m) X (i, m)], E [X2(i,m)]、E[X2(i-1, m)] be respectively the i-th frame in m-th of subband and
The mean-square value of (i-1)-th frame, E [X (i, m) X (i-1, m)] indicate the cross-correlation function of the i-th frame and the (i-1)-th frame；
S33, the pure voice signal power P for solving the i-th and (i-1)-th frame in adjacent i.e. m-th of the subband of two frames voice signal
Noise variance σ in (i, m), P (i-1, m) and m-th of subband2(m), for estimating the noise of the i-th frame of m-th of subband
Compared estimate value SNR (i, m).
Further, in step S31, the autocorrelation matrix for calculating adjacent signals vector is as follows:
Wherein, adjacent signals vector Xadjoin(i, k)={ X (i, k), X (i-1, k) }T, E (X2(i, k)), E (X2(i-1,
K)) be respectively the i-th frame and the (i-1)-th frame in k-th of Frequency point mean-square value, E (X (i, k) X (i-1, k)) indicates the i-th frame and the
The cross-correlation function of i-1 frame.
Further, in step S32, after filter obtains M subband to signal by analysis, include according to each subband
L=N/M frequency spectrum estimates the correlation matrix R (i, m) of m-th of subband consecutive frame:
In formula (2), E [X2(i,m)]、E[X2(i-1, m)] it is respectively the square of the i-th frame in m-th of subband and the (i-1)-th frame
Value, E [X (i, m) X (i-1, m)] indicate the cross-correlation function of the i-th frame and the (i-1)-th frame.
Further, in step S4, gain function is
gdB(i, m)=λim(SNR)f(xdB(n)) (13)
Wherein, gdB(i, m) represents the gain of the i-th frame voice signal in m subband.
Further, in step S4, by the gain g of the i-th frame voice signal in m subbanddB(i, m) transforms to amplitude domain g
(i, m):
The beneficial effects of the present invention are: this kind is led to based on the digital deaf-aid noise-reduction method for improving subband signal-to-noise ratio (SNR) estimation
The signal-to-noise ratio (SNR) estimation to subband signal is crossed, and calculates the yield value of each subband according to subband state of signal-to-noise, is realized to quiet
The signals and associated noises of signal or high RST noise ratio only carry out slight noise abatement, and have noise cancellation signal to improve noise abatement low signal noise ratio
Degree, alleviates the auditory fatigue of user, and solve prior art because computation complexity height due to be difficult to be suitable for pair
The difficulty of real-time and the very high hearing aid of power consumption requirements.This kind is based on the digital deaf-aid noise reduction for improving subband signal-to-noise ratio (SNR) estimation
Method, more accurately subband signal-noise ratio estimation method makes ambient noise inhibitory effect more preferable, mitigates listening for hearing aid user
Feel fatigue；Method is simple and efficient, and avoids inverse Fourier transform (IFFT) and time delay performance is largely changed
Kind, more traditional spectrum-subtraction improves 60.6%, and more traditional Wiener Filter Method improves 40.7%.
Detailed description of the invention
Fig. 1 is process signal of the embodiment of the present invention based on the digital deaf-aid noise-reduction method for improving subband signal-to-noise ratio (SNR) estimation
Figure；
Fig. 2 is that digital deaf-aid noise reduction model illustrates schematic diagram in embodiment；
Fig. 3 is modified linear-proportional gain's function in embodiment；
Fig. 4 is each algorithm delay performance comparison in embodiment.
Specific embodiment
The preferred embodiment that the invention will now be described in detail with reference to the accompanying drawings.
Digital deaf-aid mostly uses greatly multiple-channels algorithm at present, voice signal is divided into several subbands according to frequency, each
Subband individually carries out noise suppressed.Wherein according to the difference of frequency band division methods, wide and non-wide two methods can be divided into.Cause
It is different for perception of sound ability of the human ear to different frequency range, therefore generally uses non-wide frequency band division methods, it can be more
The auditory properties of good simulation human ear cochlea.In addition the filter group of threshold sampling showed in this performance of stopband attenuation compared with
Difference, therefore generally more preferable using noise suppressed performance, the over-sampling filter group of flexible design.The frequency band division methods of embodiment are adopted
It is non-wide frequency band division methods, in addition filter group uses over-sampling filter.
Embodiment
A kind of digital deaf-aid noise-reduction method based on improvement subband signal-to-noise ratio (SNR) estimation, such as Fig. 1 are decomposed using 6 rank IIR and are filtered
Original signal is resolved into 16 subbands by wave device group, then calculates in each subband the cross-correlation function of adjacent two frame signal and each
From mean-square value, to estimate the signal-to-noise ratio of the subband；Secondly, according to each subband gain of the signal-to-noise ratio computation estimated, and with
Subband signal is multiplied to obtain modified subband signal；Finally, after revised each subband signal synthesis is obtained noise reduction process
Voice.
Specifically includes the following steps:
S1, the microphone input signal x (n) of digital deaf-aid is AD converted, and the digital signal after conversion is divided
Frame xi(n) (i=0,1,2 ... N-1)；
S2, pass through resolution filter: every frame signal is passed through to 6 rank IIR resolution filter group h of multichanneli(n) (i=0,
1 ... M-1), M is port number, passband ripple 0.5dB, with ciIt is down-sampled for the progress of down-sampled coefficient, obtain M subband signal: x
(i, m) (i=0,1,2 ... N-1) (m=0,1,2 ... M-1)；
S3, subband signal-to-noise ratio (SNR) estimation: the signal-to-noise ratio (SNR) estimation value of m-th of the i-th frame of subband is
In formula, P (i, m), P (i-1, m) are the pure voice signal power of the i-th and (i-1)-th frame in m-th of subband, σ2(m) it is
Noise variance in m-th of subband, α are smoothing factor, take 0.5 in the present embodiment.
S4, yield value calculate: bringing the Signal to Noise Ratio (SNR) (i, m) estimated into gain function gdBIn (i, m), and by gdB
(i, m) transforms to amplitude domain g (i, m).
S5, subband noise inhibit: carrying out noise attentuation to primary speech signal by the yield value of obtained each subband, obtain
To enhanced voice signal: yim(n)=g (i, m) x (i, m)；
The comprehensive output of S6, subband signal: by the output signal y in M channelim(n) (i=0,1 ... M-1) first with ciTo rise
Downsampling factor carries out a liter sampling, is finally synthesizing the voice signal y (n) exported.
Embodiment based on improve subband signal-to-noise ratio (SNR) estimation digital deaf-aid noise-reduction method as shown in Fig. 2, Fig. 2 number
M is total number of channels in Noise reduction model, and embodiment divides frequency band according to auditory perceptual characteristic and COCHLEAR FILTER characteristic
For 16 subbands:
Wherein, fiIndicate the lower sideband frequencies of the i-th subband,Indicate that the upper side band frequency of the i-th subband calculates each subband
Lower sideband frequency, calculated result is as shown in table 1.
1 frequency band division result of table
In Fig. 2, x0(n),x1(n),…xM-1It (n) is by hi(n) effect is M subband for decomposing original signal
Signal, and original sampling frequency is kept, hi(n) also has the function of frequency overlapped-resistable filter simultaneously, embodiment is cut using the simulation of 3 ranks
6 rank IIR resolution filter h are designed than snow husband I typei(n) (i=0,1 ... 15), passband ripple 0.5dB.ciFor the i-th subband
Drop/liter downsampling factor.As downsampling factor ciWhen=M, referred to as threshold sampling；ciWhen < M, referred to as over-sampling.The present embodiment uses
Be over-sampling.u0(n),u1(n),…uM-1(n) for respectively by with c0(n),c1(n),…cM-1(n) it is carried out for downsampling factor
Down-sampled M obtained subband signal, down-sampled coefficient are as shown in table 2.；
The down-sampled coefficient of table 2
G in Fig. 20(n),g1(n),…gM-1(n) be each subband amount of gain, respectively with u0(n),u1(n),…uM-1(n) phase
The multiplied output signal v to after sub--band speech enhances algorithm process0(n),v1(n),…vM-1(n)；Treated v0(n),v1
(n),…vM-1(n) signal is using with c0(n),c1(n),…cM-1(n) a liter sampling is carried out for downsampling factor.Final each subband
Signal synthesis, the clean speech signal of output estimation.
Subband signal-to-noise ratio (SNR) estimation
Since voice signal and noise are time varying signal, in each frame, the frequency domain distribution of voice and noise energy is
It is unbalanced, it is therefore desirable to the signal-to-noise ratio of real-time estimation signal.The present invention proposes a kind of improved signal-noise ratio estimation method, utilizes
The autocorrelation matrix of adjacent two frames noisy speech signal, estimates the signal-to-noise ratio of each subband.
Subband signal-to-noise ratio (SNR) estimation is specific as follows:
Assuming that the i-th frame signal are as follows:
X (i, n)=s (i, n)+n (i, n) (2)
In formula, x (i, n) represents Noisy Speech Signal, and s (i, n) represents primary speech signal, and n (i, n) represents noise letter
Number.
Its Fourier transformation are as follows:
X (i, k)=S (i, k)+N (i, k) (3)
Defining the i-th and (i-1)-th frame signal on kth Frequency point is adjacent signals vector Xadjoin(i, k), then
Xadjoin(i, k)={ X (i, k), X (i-1, k) }T
Calculate the autocorrelation matrix of adjacent signals vector:
If only estimating the correlation matrix of each frequency point with X (i, k) and X (i-1, k), larger evaluated error can be generated.
It can include L according to each subband after filter obtains M subband to signal by analysismA frequency spectrum estimates m-th of subband
The correlation matrix R (i, m) of consecutive frame:
In formula (6), E [X2(i,m)]、E[X2(i-1, m)] it is respectively the square of the i-th frame in m-th of subband and the (i-1)-th frame
Value, E [X (i, m) X (i-1, m)] indicate the cross-correlation function of the i-th frame and the (i-1)-th frame.
Assuming that the clean speech signal s (t) of any frame is unrelated with the noise signal of any frame statistics, while consecutive frame is made an uproar
Sound is uncorrelated, then the items in formula (6) can abbreviation be
E[X2(i, m)]=P (i, m)+σ2(m) (6)
E[X2(i-1, m)]=P (i-1, m)+σ2(m) (7)
Wherein, P (i, m), P (i-1, m) are the pure voice signal power of the i-th and (i-1)-th frame in m-th of subband, σ2(m) it is
Noise variance in m-th of subband.
E [X (i, m) X (i-1, m)]=E [S (i, m) S (i-1, m)] (7+)
Above formula is deformed, can be obtained
From the above equation, we can see that
λ (i-1, m)=S (i, m)/S (i-1, m)=(X (i, m)-N (i, m))/(X (i-1, m)-N (i-1, m)) (8+)
Assuming that noise is smooth change, the noise spectrum N that N (i, m) and N (i-1, m) estimate when being mutesilence(m)。
Such R (i, m) may finally be expressed as
Utilize the L in each subbandmA frequency spectrum estimates the E [X in (7) (8) (9) formula2(i, m)], E [X2(i-1,
M)], E [X (i-1, m) X (i, m)].
Bring into formula (7), (8), in (9), P (i, m), σ2(m), P (i-1, m) can be uniquely determined.
The signal-to-noise ratio (SNR) estimation value of such the i-th frame of m-th of subband is
The construction of gain function
The basic thought of Wiener filtering noise reduction model is by designing a linear filter H (w), so that linear by this
Output after filter can reach mean square error expectation minimum, i.e.,For the optimal estimation of clean speech S (w).
Wherein,PsIt (w) is clean speech signal power spectral density, PdIt (w) is noise power spectrum
Density.Noise estimation generally uses the noise spectrum estimation algorithm based on prior weight and posteriori SNR, and flat by introducing
Sliding parameter exports filter transfer function H (w).
Wiener Filter Method be by being multiplied by a gain function to noisy speech or by a filter, the purpose is to
In order to by the sound pressure level of noise attentuation to patient comfort.Therefore embodiment contains quiet signal or high RST noise ratio to realize
Noise cancellation signal carries out slight noise abatement without noise abatement or only, and improves reduction to the noise cancellation signal that has of low signal noise ratio, will increase
Beneficial function is defined as
gdB(i, m)=λim(SNR)f(xdB(n)) (15)
Wherein gdB(i, m) represents the gain (domain dB) of the i-th frame voice signal in m subband, and g (i, m) is by gdB(i,m)
Transform to amplitude domain.λim(SNR) pad value that the i-th frame voice signal changes with SNR in m subband is represented, by λim(SNR) it limits
System is in [- 1,0].Work as λim(SNR)=0 when, amount of gain is 1 after being transformed into amplitude domain, is represented undamped；λim(SNR) it more connects
When closely -1, the pad value in respective amplitude domain is bigger.
Although existing Noise reduction algorithm utilizes pad value function lambdaim(SNR) noise reduction is carried out with the relationship of subband SNR,
But under low signal-to-noise ratio, ambient noise is still remained excessively, will lead to user's auditory fatigue.For this purpose, embodiment proposes one kind
Improvement strategy, in low signal-to-noise ratio section modified gain proportion function, as shown in figure 3, low signal-to-noise ratio section [0, B0] attenuation curve is precipitous,
Purpose is ambient noise when significantly weakening low signal-to-noise ratio.But this strategy be for faint quiet signal it is unfavorable, because
This gain function should also depend on environmental noise level.When being in quietly interior such as patient, if relying solely on signal noise ratio
Calculating gain will lead to faint quiet signal and significantly be attenuated, and introduce maximum noise attenuation function f (N thusdB(n)) to control
Signal attenuation amplitude processed.In this way under quiet environment, NdB(n) very low, maximum attenuation amplitude f (N is setdB(n))=NdB(n)
Afterwards, it is almost not required to decay signal noise ratio is very low.
Performance test analysis
The modulation depth method that algorithm proposed by the present invention and patent US2004/6757395 B1 are proposed is compared by testing
Algorithm performance difference, and compare with basic spectrum-subtraction, Wiener Filter Method the performance of processing delay.Test all clean speech
It is derived from TIMIT speech database, sample rate 16kHz.From Noise92 noise library, noise type is respectively for pure noise interception
White, Pink, Tank and Speech Babble, input signal-to-noise ratio take 0dB, 5dB, 10dB.
Speech perceptual quality evaluation
Embodiment is evaluated using logarithm spectral measure LSD (Log-spectral distortion) and speech perceptual quality
PESQ (Perceptual evaluation of speech quality) is objective evaluation index.
LSD reflection is voice distortion situation, and PESQ will reflect voice quality on the whole.Both indexs are commented with subjectivity
The valence degree of correlation with higher.General LSD drop-out value (Δ LSD) is bigger, shows that the logarithmic spectrum distortion factor is smaller, algorithm is to voice
Degree of injury is smaller, and PESQ raising amount (Δ PESQ) is bigger, and the voice quality after showing algorithm process is better.LSD and PESQ
Calculation method it is as follows
PESQ=4.5-0.1dSYM-0.0309·dASYM (18)
In formula (17), X (k, l) andThe respectively Short Time Fourier Transform of clean speech and enhanced voice, N
For frame length, J is frame number.D in formula (18)SYMAnd dASYMFor the correlometer formula in cognitive model.Calculated result is as shown in table 3.
The evaluation comparison of the speech perceptual quality of 3 modulation depth method of table and the method for the present invention
In table 3, comparing input signal-to-noise ratio is respectively 0dB, when 5dB, 10dB, side that modulation depth method and embodiment are proposed
The raising amount of LSD slippage and PESQ of the method under different noise types, it can be found that the method that proposes of embodiment in White and
It is excellent under Pink noise circumstance, the improvement degree of LSD and PESQ are substantially better than modulation depth method, especially make an uproar in White
Under acoustic environment, the method for the present invention is averagely higher by 36.7%, PESQ improved values in logarithmic spectrum compared with modulation depth method in improved values
Averagely it is higher by 19%.But under Tank and Speech Babble noise circumstance, the present invention and modulation depth method performance are showed
Bad, the average improvement amount of PESQ is only 0.1, so linear scale attenuation model used in the present invention is further improved,
Be not suitable for Tank and Speech Babble noise circumstance, therefore the performance of subjective feeling is influenced by noise scenarios.
Algorithm complexity and time delay are analyzed
The noisy speech duration 2.26s wherein handled, experiment frame length take 64,128 and 256 respectively, and frame shifting is 50%.Though
So multiband algorithm is both needed to signal is divided into several subbands by analysis filter, but the calculation amount of filter equalizer
It is not belonging in the range of present invention consideration.Therefore, the processing delay that embodiment is counted is passed through after resolution filter from signal
Process before to synthesis filter.As can be known from Fig. 4, the time delay of basic spectrum-subtraction and basic Wiener Filter Method will be significantly greater than
Method proposed by the invention, this is because these two types of models are both needed to by Fast Fourier Transform (FFT) (Fast Fourier
Transform, FFT) frequency-domain analysis is carried out, and finally by inverse fast Fourier transform (Inverse Fast Fourier
Transform, IFFT) time-domain signal is reverted to, therefore calculation amount can dramatically increase.And the present invention estimating in subband signal-to-noise ratio
Timing carries out frequency-domain analysis by means of FFT, and the calculating of FFT is only used for estimator (13), the cross-correlation function being directed to and
The calculation amount being just worth all is extremely low.In addition the present invention is not required to revert to the process of time-domain signal via IFFT.Because of the invention
Computation complexity is lower than basic spectrum-subtraction and basic Wiener Filter Method.Frame length 2 is usually taken in engineering8It is=256 points, of the invention at this time
The more basic spectrum-subtraction of processing delay performance improve 60.6%, improve 40.7% compared with Wiener Filter Method.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, without departing from the technical principles of the invention, several improvement and deformations can also be made, these improvement and deformations
Also it should be regarded as protection scope of the present invention.
Claims (4)
1. a kind of based on the digital deaf-aid noise-reduction method for improving subband signal-to-noise ratio (SNR) estimation characterized by comprising
Original signal is resolved by several subbands using resolution filter group, then calculates adjacent two frame signal in each subband
Cross-correlation function and respective mean-square value, to estimate the signal-to-noise ratio of the subband；
Secondly, according to each subband gain of the signal-to-noise ratio computation estimated, and be multiplied to obtain modified subband signal with subband signal；
Finally, revised each subband signal synthesis to be obtained to the voice after noise reduction process；
Specifically includes the following steps:
S1, the microphone input signal x (n) of digital deaf-aid is AD converted, and by the digital signal framing x after conversioni
(n), wherein i=0,1,2 ... N-1；
S2, pass through resolution filter: every frame signal is passed through to 6 rank IIR resolution filter group h of multichanneli(n), wherein i=0,
1 ... M-1, M are port number, with ciIt is down-sampled for the progress of down-sampled coefficient, obtain M subband signal: x (i, m), wherein i=
0,1,2 ... N-1, m=0,1 ... M-1；
S3, subband signal-to-noise ratio (SNR) estimation: the signal-to-noise ratio (SNR) estimation value of m-th of the i-th frame of subband is
In formula, P (i, m), P (i-1, m) are the pure voice signal power of the i-th and (i-1)-th frame in m-th of subband, σ2(m) it is m-th
Noise variance in subband, α are smoothing factor, 0 < α < 1；
S4, yield value calculate: bringing the Signal to Noise Ratio (SNR) (i, m) estimated into gain function gdBIn (i, m), and by gdB(i, m) becomes
Change to amplitude domain g (i, m)；In step S4, by the gain g of the i-th frame voice signal in m subbanddB(i, m) transforms to amplitude domain g
(i, m):
S5, subband noise inhibit: carrying out noise attentuation to primary speech signal by the yield value of obtained each subband, increased
Voice signal after strong: yim(n)=g (i, m) x (i, m)；
The comprehensive output of S6, subband signal: by the output signal y in M channelim(n), wherein i=0,1 ... M-1, first with ciTo rise
Downsampling factor carries out a liter sampling, is finally synthesizing the voice signal y (n) exported.
2. as described in claim 1 based on the digital deaf-aid noise-reduction method for improving subband signal-to-noise ratio (SNR) estimation, which is characterized in that
In step S3, subband signal-to-noise ratio (SNR) estimation specifically:
S31, the L in each subband is utilizedmA frequency spectrum estimates the E [X in correlation matrix R (i, m)2(i,m)]、E[X2(i-1,
] and E [X (i-1, m) X (i, m)], E [X m)2(i,m)]、E[X2(i-1, m)] it is respectively the i-th frame and (i-1)-th in m-th of subband
The mean-square value of frame, E [X (i, m) X (i-1, m)] indicate the cross-correlation function of the i-th frame and the (i-1)-th frame；
S32, solve the i-th and (i-1)-th frame in adjacent i.e. m-th of the subband of two frames voice signal pure voice signal power P (i,
M), the noise variance σ in P (i-1, m) and m-th of subband2(m), estimate for estimating the signal-to-noise ratio of the i-th frame of m-th of subband
Evaluation SNR (i, m).
3. as claimed in claim 2 based on the digital deaf-aid noise-reduction method for improving subband signal-to-noise ratio (SNR) estimation, which is characterized in that
It include L according to each subband after filter obtains M subband to signal by analysis in step S31mA frequency spectrum estimates m
The correlation matrix R (i, m) of a subband consecutive frame:
In formula (2), E [X2(i,m)]、E[X2(i-1, m)] be respectively the i-th frame and the (i-1)-th frame in m-th of subband mean-square value, E
The cross-correlation function of [X (i, m) X (i-1, m)] expression the i-th frame and the (i-1)-th frame.
4. the digital deaf-aid noise-reduction method as described in any one of claims 1-3 based on improvement subband signal-to-noise ratio (SNR) estimation,
It is characterized in that, in step S4, gain function is
gdB(i, m)=λim(SNR)f(xdB(n)) (13)
Wherein, gdB(i, m) represents the gain of the i-th frame voice signal in m subband, λimIt (SNR) is the i-th frame voice in m subband
The pad value that signal changes with SNR, by λim(SNR) it is limited in [- 1,0], f (xdBIt (n)) is maximum noise attenuation function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610150663.0A CN105679330B (en) | 2016-03-16 | 2016-03-16 | Based on the digital deaf-aid noise-reduction method for improving subband signal-to-noise ratio (SNR) estimation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610150663.0A CN105679330B (en) | 2016-03-16 | 2016-03-16 | Based on the digital deaf-aid noise-reduction method for improving subband signal-to-noise ratio (SNR) estimation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105679330A CN105679330A (en) | 2016-06-15 |
| CN105679330B true CN105679330B (en) | 2019-11-29 |
Family
ID=56310634
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610150663.0A Active CN105679330B (en) | 2016-03-16 | 2016-03-16 | Based on the digital deaf-aid noise-reduction method for improving subband signal-to-noise ratio (SNR) estimation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105679330B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106301553B (en) * | 2016-08-15 | 2018-07-06 | 北京邮电大学 | Determine interferometer both arms time delayed difference value method, OSNR Monitoring Method and device |
| CN107680610A (en) * | 2017-09-27 | 2018-02-09 | 安徽硕威智能科技有限公司 | A kind of speech-enhancement system and method |
| CN107889006B (en) * | 2017-10-11 | 2019-08-27 | 恒玄科技（上海）有限公司 | A kind of active noise reduction system of flexible modulation de-noising signal delay |
| CN108257617B (en) * | 2018-01-11 | 2021-01-19 | 会听声学科技（北京）有限公司 | Noise scene recognition system and method |
| CN108281154B (en) * | 2018-01-24 | 2021-05-18 | 成都创信特电子技术有限公司 | Noise reduction method for voice signal |
| CN110473567B (en) * | 2019-09-06 | 2021-09-14 | 上海又为智能科技有限公司 | Audio processing method and device based on deep neural network and storage medium |
| CN112885375A (en) * | 2021-01-08 | 2021-06-01 | 天津大学 | Global signal-to-noise ratio estimation method based on auditory filter bank and convolutional neural network |
| CN115604627A (en) * | 2022-10-25 | 2023-01-13 | 维沃移动通信有限公司（Cn） | Audio signal processing method and device, electronic equipment and readable storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101582264A (en) * | 2009-06-12 | 2009-11-18 | 瑞声声学科技(深圳)有限公司 | Method and voice collecting system for speech enhancement |
| CN102004264A (en) * | 2010-10-18 | 2011-04-06 | 中国石油化工股份有限公司 | Quantitative analysis and evaluation method for quality of acquired seismic data |
| CN103607363A (en) * | 2013-12-04 | 2014-02-26 | 北京邮电大学 | Blind estimation method of signal to noise ratio |
| CN104917706A (en) * | 2014-03-10 | 2015-09-16 | 联想(北京)有限公司 | Signal-to-Noise Ratio estimation method and electronic equipment |
| CN105390142A (en) * | 2015-12-17 | 2016-03-09 | 广州大学 | Digital hearing aid voice noise elimination method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8988970B2 (en) * | 2010-03-12 | 2015-03-24 | University Of Maryland | Method and system for dereverberation of signals propagating in reverberative environments |
-
2016
- 2016-03-16 CN CN201610150663.0A patent/CN105679330B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101582264A (en) * | 2009-06-12 | 2009-11-18 | 瑞声声学科技(深圳)有限公司 | Method and voice collecting system for speech enhancement |
| CN102004264A (en) * | 2010-10-18 | 2011-04-06 | 中国石油化工股份有限公司 | Quantitative analysis and evaluation method for quality of acquired seismic data |
| CN103607363A (en) * | 2013-12-04 | 2014-02-26 | 北京邮电大学 | Blind estimation method of signal to noise ratio |
| CN104917706A (en) * | 2014-03-10 | 2015-09-16 | 联想(北京)有限公司 | Signal-to-Noise Ratio estimation method and electronic equipment |
| CN105390142A (en) * | 2015-12-17 | 2016-03-09 | 广州大学 | Digital hearing aid voice noise elimination method |
Non-Patent Citations (1)
| Title |
|---|
| 地震记录信噪比估算方法研究;王红玲;《中国优秀硕士学位论文全文数据库基础科学辑》;20071215;第2章第4-9页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105679330A (en) | 2016-06-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105679330B (en) | Based on the digital deaf-aid noise-reduction method for improving subband signal-to-noise ratio (SNR) estimation | |
| CN105741849B (en) | The sound enhancement method of phase estimation and human hearing characteristic is merged in digital deaf-aid | |
| CN110085249B (en) | Single-channel speech enhancement method of recurrent neural network based on attention gating | |
| Wu et al. | A two-stage algorithm for one-microphone reverberant speech enhancement | |
| CN108447495B (en) | Deep learning voice enhancement method based on comprehensive feature set | |
| CN101901602B (en) | Method for reducing noise by using hearing threshold of impaired hearing | |
| CN105390142B (en) | A kind of digital deaf-aid voice noise removing method | |
| WO2011137258A1 (en) | Multi-microphone robust noise suppression | |
| CN108986832B (en) | Binaural voice dereverberation method and device based on voice occurrence probability and consistency | |
| Ren et al. | A Causal U-Net Based Neural Beamforming Network for Real-Time Multi-Channel Speech Enhancement. | |
| CN109961799A (en) | A kind of hearing aid multicenter voice enhancing algorithm based on Iterative Wiener Filtering | |
| Shi et al. | Robust speaker recognition based on improved GFCC | |
| Taal et al. | SII-Based Speech Prepocessing for Intelligibility Improvement in Noise | |
| Bhat et al. | Smartphone based real-time super gaussian single microphone speech enhancement to improve intelligibility for hearing aid users using formant information | |
| Inoue et al. | Theoretical analysis of musical noise in Wiener filtering family via higher-order statistics | |
| CN105869649A (en) | Perceptual filtering method and perceptual filter | |
| CN117219102A (en) | Low-complexity voice enhancement method based on auditory perception | |
| Hsu et al. | Spectro-temporal subband wiener filter for speech enhancement | |
| Rao et al. | Speech enhancement using sub-band cross-correlation compensated Wiener filter combined with harmonic regeneration | |
| CN114566179A (en) | Time delay controllable voice noise reduction method | |
| Miyazaki et al. | Theoretical analysis of parametric blind spatial subtraction array and its application to speech recognition performance prediction | |
| Gui et al. | Adaptive subband Wiener filtering for speech enhancement using critical-band gammatone filterbank | |
| Sun et al. | An RNN-based speech enhancement method for a binaural hearing aid system | |
| Surendran et al. | Perceptual subspace speech enhancement with variance normalization | |
| Ngo et al. | A flexible speech distortion weighted multi-channel Wiener filter for noise reduction in hearing aids |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |