CN105931648A - Audio signal de-reverberation method and device - Google Patents
Audio signal de-reverberation method and device Download PDFInfo
- Publication number
- CN105931648A CN105931648A CN201610474006.1A CN201610474006A CN105931648A CN 105931648 A CN105931648 A CN 105931648A CN 201610474006 A CN201610474006 A CN 201610474006A CN 105931648 A CN105931648 A CN 105931648A
- Authority
- CN
- China
- Prior art keywords
- filter
- signal
- audio signal
- sub
- channel audio
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 230000005236 sound signal Effects 0.000 title claims abstract description 136
- 238000000034 method Methods 0.000 title claims abstract description 54
- 230000003595 spectral effect Effects 0.000 claims description 25
- 230000017105 transposition Effects 0.000 claims description 20
- 238000001914 filtration Methods 0.000 claims description 15
- 239000011159 matrix material Substances 0.000 claims description 10
- 238000005516 engineering process Methods 0.000 claims description 9
- 230000000694 effects Effects 0.000 claims description 8
- 238000001514 detection method Methods 0.000 claims description 7
- 206010027336 Menstruation delayed Diseases 0.000 claims 1
- 238000001228 spectrum Methods 0.000 claims 1
- 230000008569 process Effects 0.000 description 16
- 230000006854 communication Effects 0.000 description 7
- 238000004891 communication Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 5
- 238000004590 computer program Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 230000033228 biological regulation Effects 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 241001062009 Indigofera Species 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 239000011358 absorbing material Substances 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000010924 continuous production Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000002427 irreversible effect Effects 0.000 description 1
- 210000003127 knee Anatomy 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 239000004065 semiconductor Substances 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
- G10L19/0204—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 subband 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
- G10L2021/02082—Noise filtering the noise being echo, reverberation of the speech
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Quality & Reliability (AREA)
- Stereophonic System (AREA)
- Cable Transmission Systems, Equalization Of Radio And Reduction Of Echo (AREA)
Abstract
The invention discloses an audio signal de-reverberation method and device. The method comprises the steps that a single channel audio signal is acquired, wherein the single channel audio signal comprises an early reverberation signal and a late reverberation signal; whether the single channel audio signal is a speech signal is determined; if so, the variance of the joint probability density distribution of the early reverberation signal is updated, and the filter coefficient of a subband filter is updated based on the variance of the joint probability density distribution of the early reverberation signal, wherein the subband filter is used to filter out the late reverberation signal in the single channel audio signal; and the de-reverberation single channel audio signal is determined based on the updated filter coefficient. According to the scheme provided by the invention, the late reverberation signal in the inputted audio signal can be filtered out to improve the accuracy of subsequent speech recognition.
Description
Technical field
The application relates to field of computer technology, is specifically related to Audio Signal Processing field, especially
Relate to audio signal solution reverberation method and device.
Background technology
During audio frequency especially Speech processing, if for gathering adopting of audio signal
Acquisition means (such as, mike) distance sound source position farther out, the sound that harvester receives
Frequently signal will inevitably be affected by reverberation.The existence of reverberation not only can reduce audio frequency letter
Number sense of hearing quality, and the precision of existing voice identification system can be caused drastically to decline.
Reverberation can be decomposed into reverberation in early days and late reverberation, wherein to audio quality and identification
What system accuracy impact was bigger is late reverberation, and the main target therefore solving reverberation is how to reduce
Late reverberation.
In the prior art, some have been there are for filtering the voice signal that harvester collects
In the algorithm of late reverberation.But, the most all there is following problem in these algorithms:
1) when solving for the filter coefficient of the wave filter filtering late reverberation, need to obtain
Whole voice data, thus cause the real-time rate of algorithm relatively low, and then can cause solving reverberation and calculate
The time delay of method is higher.And in voice communication and field of speech recognition, to solving reverberation algorithm
Requirement of real-time is higher.
2), when solving for the filter coefficient of the wave filter filtering late reverberation, it will usually
Relate to the calculating of matrix inversion.And once matrix is irreversible during matrix inversion, then try to achieve
Filter coefficient the most inaccurate, and then impact solve reverberation performance.Additionally, matrix inversion
Operand is relatively big, and the real-time also resulting in understanding reverberation algorithm from another point of view is poor.
Summary of the invention
The purpose of the application is to propose the audio signal solution reverberation method of a kind of improvement and device,
Solve the technical problem that background section above is mentioned.
First aspect, this application provides a kind of audio signal solution reverberation method, including: obtain
Channel audio signal, channel audio signal includes reverb signal and late reverberation signal in early days；
Judge whether channel audio signal is voice signal；The most then update early stage reverb signal
The variance of joint probability density distribution, and joint probability density based on early stage reverb signal distribution
Variance update sub-filter filter coefficient, wherein, sub-filter is used for filtering list
The late reverberation signal comprised in channel audio signal；And based on the filter coefficient after updating
Determine the audio signal after solving reverberation.
In certain embodiments, it is judged that whether channel audio signal is that voice signal includes: logical
Cross voice activity detection technology and judge whether channel audio signal is voice signal.
In certain embodiments, the variance of the joint probability density distribution of reverb signal in early days is more
The transposed matrix of the filter coefficient of the sub-filter before new and the input signal of sub-filter
The product of vector and the spectral coefficient x of the channel audio signal received in tt,fDifference exhausted
To value square；Filter coefficient g after renewalf(t+1) for update before filter coefficient with more
New change amount sum；Wherein, updating variable quantity is that the first renewal running parameter and second updates change
Change the ratio of parameter；First renewal running parameter is the iteration step length of sub-filter, sub-band filter
Error between actual output and the desired output of device and the input signal of sub-filter to
The product of amount；Second update running parameter be sub-filter input signal vector transposition with
The product of the input signal vector of sub-filter；The actual output of sub-filter is defeated with expectation
Error between going out is equal to the spectral coefficient of the channel audio signal received in t and t
The ratio of variance of joint probability density distribution of early stage reverb signal deduct the subband filter before renewal
The transposition of the filter coefficient of ripple device and the product of the input signal vector of sub-filter.
In certain embodiments, audio signal d after reverberation is solvedt,fEqual to receive in t
The spectral coefficient of channel audio signal deducts the filter coefficient of the sub-filter in t+1 moment
The product of the input signal vector of transposition and sub-filter.In certain embodiments, judging
After whether channel audio signal is voice signal, method also includes: if it is not, then will update
Sub-filter before the variance of the joint probability density distribution of front early stage reverb signal and renewal
Filter coefficient as update after early stage reverb signal joint probability density distribution variance
Filter coefficient with the sub-filter after renewal.
In certain embodiments, method also includes: judge actual output and the phase of sub-filter
Hope the error between output whether meet Error Absolute Value square more than receiving in t
The absolute value of the spectral coefficient of channel audio signal square with the product of predetermined threshold value K；If so,
Then filter coefficient is set to null vector；Wherein, K ＞ 1.
Second aspect, this application provides a kind of audio signal solution reverberation unit, including: obtain
Module, is configured to obtain channel audio signal, and channel audio signal includes reverberation in early days
Signal and late reverberation signal；Judge module, is configured to whether judge channel audio signal
For voice signal；First more new module, if being configured to channel audio signal is voice signal,
Then update the variance of the joint probability density distribution of reverb signal in early days, and believe based on early stage reverberation
Number joint probability density distribution variance update sub-filter filter coefficient, wherein,
Sub-filter is for filtering the late reverberation signal comprised in channel audio signal；And really
Cover half block, is configured to determine the audio signal after solving reverberation based on the filter coefficient after updating.
In certain embodiments, it is judged that module is configured to further: pass through voice activity detection
Technology judges whether channel audio signal is voice signal.
In certain embodiments, the variance of the joint probability density distribution of reverb signal in early days is:
The input letter of the transposed matrix of the filter coefficient of the sub-filter before renewal and sub-filter
The spectral coefficient x of number vectorial product and the channel audio signal received in tt,fDifference
Absolute value square；Filter coefficient g after renewalf(t+1) for update before filter coefficient with
Update variable quantity sum；Wherein, updating variable quantity is that the first renewal running parameter and second updates
The ratio of running parameter；First renewal running parameter is the iteration step length of sub-filter, subband filter
Error between actual output and the desired output of ripple device and the input signal of sub-filter
The product of vector；Second transposition updating the input signal vector that running parameter is sub-filter
Product with the input signal vector of sub-filter；The actual output of sub-filter and expectation
Error between output is equal to the spectral coefficient of the channel audio signal received in t and t
The ratio of the variance of the joint probability density distribution of the early stage reverb signal in moment deducts the son before renewal
The transposition of the filter coefficient of band filter and the product of the input signal vector of sub-filter.
In certain embodiments, audio signal d after reverberation is solvedt,fEqual to receive in t
The spectral coefficient of channel audio signal deducts the filter coefficient of the sub-filter in t+1 moment
The product of the input signal vector of transposition and sub-filter.
In certain embodiments, device also includes the second more new module；Second more new module configuration
If for channel audio signal not being voice signal, then the early stage reverb signal before updating
The filter coefficient of the sub-filter before the variance of joint probability density distribution and renewal is as more
Sub-band filter after the variance of the joint probability density distribution of the early stage reverb signal after Xin and renewal
The filter coefficient of device.
In certain embodiments, device also includes zero setting module；Zero setting module is configured to judge
Whether the error between actual output and the desired output of sub-filter meets Error Absolute Value
Square more than the channel audio signal received in t spectral coefficient absolute value square
Product with predetermined threshold value K；And the most then filter coefficient is set to null vector；Wherein,
K ＞ 1.
The audio signal solution reverberation method of the application offer and device, by voice based on input
Signal is continuously updated variance and the subband filter of the joint probability density distribution of reverb signal in early days
The parameter of ripple device so that sub-filter filtering can be by the channel audio signal of input
Late reverberation target signal filter, thus improve the accuracy rate of subsequent speech recognition.
Additionally, the audio signal solution reverberation method of the application and device, the connection of reverb signal in early days
Renewal and the sub-filter parameter of closing the variance of probability density distribution update required amount of calculation relatively
Little, and renewal process is only relevant to the channel audio signal in a period of time, has higher
Real-time.
Accompanying drawing explanation
By reading retouching in detail with reference to made non-limiting example is made of the following drawings
Stating, other features, purpose and advantage will become more apparent upon:
Fig. 1 is that the application can apply to exemplary system architecture figure therein；
Fig. 2 is the flow process of an embodiment of the audio signal solution reverberation method according to the application
Figure；
Fig. 3 is the flow process of another embodiment of the audio signal solution reverberation method according to the application
Figure；
Fig. 4 is that the structure of an embodiment of the audio signal solution reverberation unit according to the application is shown
It is intended to；
Fig. 5 is adapted for the computer for the terminal unit or server realizing the embodiment of the present application
The structural representation of system.
Detailed description of the invention
With embodiment, the application is described in further detail below in conjunction with the accompanying drawings.It is appreciated that
, specific embodiment described herein is used only for explaining related invention, rather than to this
Bright restriction.It also should be noted that, for the ease of describe, accompanying drawing illustrate only with
About the part that invention is relevant.
It should be noted that in the case of not conflicting, the embodiment in the application and embodiment
In feature can be mutually combined.Describe this below with reference to the accompanying drawings and in conjunction with the embodiments in detail
Application.
Fig. 1 shows audio signal solution reverberation method or the audio signal solution that can apply the application
The exemplary system architecture 100 of the embodiment of reverberation unit.
As it is shown in figure 1, system architecture 100 can include terminal unit 101,102,103,
Network 104 and server 105.Network 104 is in order at terminal unit 101,102,103 and
The medium of communication link is provided between server 105.Network 104 can include various connection class
Type, the most wired, wireless communication link or fiber optic cables etc..
User can use terminal unit 101,102,103 by network 104 and server 105
Alternately, to receive or to send message etc..Can be provided with on terminal unit 101,102,103
Various telecommunication customer ends are applied, and such as web browser applications, shopping class application, searching class are answered
With, JICQ, mailbox client, social platform software etc..
Terminal unit 101,102,103 can be ability various having and gathering audio signal
Electronic equipment, includes but not limited to smart mobile phone, panel computer, MP3 player (Moving
Picture Experts Group Audio Layer III, dynamic image expert's compression standard audio layer
Face 3), (Moving Picture Experts Group Audio Layer IV, dynamic image is special for MP4
Family's compression standard audio frequency aspect 4) player, pocket computer on knee and desk computer etc..
Server 105 can be to provide the server of various service, such as to terminal unit 101,
102,103 audio signals gathered carry out the audio processing service device processed.Audio processing service
The data such as the audio signal received can be analyzed waiting and process by device, and by result (example
Voice data as processed through reverberation removal) feed back to terminal unit.
It should be noted that the audio signal solution reverberation method that the embodiment of the present application is provided both may be used
To be performed by terminal unit 101,102,103, it is also possible to performed by server 105, also may be used
With a part of step by terminal unit 101,102,103 perform and another part step by servicing
Device 105 performs.Correspondingly, audio signal solution reverberation unit both can be arranged at terminal unit 101,
102, in 103, it is also possible to be arranged in server 105, or a part of module is arranged at end
In end equipment 101,102,103, another part module is arranged in server.
It should be understood that the number of terminal unit, network and the server in Fig. 1 is only signal
Property.According to realizing needs, can have any number of terminal unit, network and server.
With continued reference to Fig. 2, it is shown that according to one of the audio signal solution reverberation method of the application
The flow process 200 of embodiment.Described audio signal solution reverberation method, comprises the following steps:
Step 210, obtains channel audio signal, and wherein, channel audio signal includes early
Phase reverb signal and late reverberation signal.
In the present embodiment, audio signal solution reverberation method runs on electronic equipment (example thereon
Server as shown in Figure 1) can by wired connection mode or radio connection from
Family utilizes it to carry out the terminal of audio signal sample and receives channel audio signal.It may be noted that
It is that above-mentioned radio connection can include but not limited to that 3G/4G connects, WiFi connects, indigo plant
Tooth connects, WiMAX connects, Zigbee connects, UWB (ultra wideband) connects,
And other currently known or exploitation in the future radio connection.
During it is said that in general, there is a certain distance between equipment and the sound source gathering sound, adopt
The audio signal that the equipment of collection sound collects will be affected by reverberation.Reverberation is generally according to directly
Reach sound (namely directly being conveyed to gather the audio signal of the equipment of sound from sound source) and be transferred to
Time difference between the reverberation of the equipment gathering sound is reverberation in early days and late reverberation.Such as,
Arrival in 30ms (millisecond) after direct sound wave can be gathered the reverberation of equipment of sound as in early days
Reverberation, and the reverberation of the equipment gathering sound will be arrived more than 30ms as late reverberation.Early
Voice amplitude, phase delay, formant are affected less by phase reverberation, and late reverberation then can be right
The impact of voice amplitude, phase delay, formant is bigger, and can cause the phase of linguistic syllables
Inter-masking, these all reduce speech intelligibility, bring the biggest difficulty to speech recognition.
Step 220, it is judged that whether channel audio signal is voice signal.
The audio signal solution reverberation method of the present embodiment, it is intended to the voice signal collected is carried out
Solution reverberation processes, and then the voice signal after processing can be reflected more realistically, and user is actual
The voice said.And then promote the accuracy rate of the signal processing such as subsequent speech recognition.
Additionally, voice signal and other audio signal exist more significantly difference, therefore, logical
Cross in step, judge whether the channel audio signal got is voice signal, can be rear
In continuous process step, carry out the operations such as corresponding filtering only for voice signal, thus improve
The solution reverberation treatment effeciency of voice signal and real-time.
In some optional implementations, for example, it is possible to by VAD (Voice Activity
Detection, voice activity detection) technology carries out the identification of voice signal, thus judges
Whether the channel audio signal got in step 210 is voice signal.
Step 230, the most then update the side of the joint probability density distribution of reverb signal in early days
Difference, and the variance renewal sub-filter of joint probability density based on early stage reverb signal distribution
Filter coefficient.
The variance of the joint probability density distribution of reverb signal can be the subband filter before updating in early days
The product of the transposed matrix of the filter coefficient of ripple device and the input signal vector of sub-filter with
Spectral coefficient x at the channel audio signal that t receivest,fDifference absolute value square.
Exemplarily, the variance of the joint probability density distribution of reverb signal in early days can meet:
Wherein, xt,fThe spectral coefficient of the channel audio signal for receiving in t,For more
The transposition of the filter coefficient of the sub-filter before new,Input for sub-filter is believed
Number vector.
Filter coefficient g after renewalf(t+1) for the filter coefficient before updating and renewal variable quantity
Sum.Wherein, update variable quantity be the first renewal running parameter and second update running parameter it
Ratio.First updates the reality that running parameter is the iteration step length of sub-filter, sub-filter
Error between output and desired output and the product of the input signal vector of sub-filter.
Second transposition updating the input signal vector that running parameter is sub-filter and sub-filter
The product of input signal vector.Mistake between actual output and the desired output of sub-filter
The poor spectral coefficient equal to the channel audio signal received in t and the early stage reverberation of t
The ratio of the variance of the joint probability density distribution of signal deducts the filtering of the sub-filter before renewal
The transposition of device coefficient and the product of the input signal vector of sub-filter.
Exemplarily, the filter coefficient g after renewalf(t+1) can meet:
Wherein, μ is the iteration step length of sub-filter, and e (t) is the actual output of sub-filter
And the error e (t) between desired output meets:
Here, the initial value g of the filter coefficient of sub-filterf(0) can be that single order number is
The full null vector of N.
Wherein, N is the number of taps of sub-filter, and meets:
N=L-D+1.
L is the frame time length ratio of reverberation time and channel audio signal one frame, and D is early
The reverberation time of phase reverberation and the frame time length ratio of channel audio signal one frame.And reverberation
Time such as can be defined as indoor sound and reach steady statue, and sound source is remaining sound after stopping sounding
Sound absorbs through sound-absorbing material in room repeatedly, the time needed for average sound pressure level decay 60dB.
Additionally, the input signal vector of sub-filterCan have a following form of expression:
In other words,Can be understood as sub-filter to connect to the t-L+1 moment from the t-D moment
The input signal vector that the input signal received is formed.
Additionally, in formula (3):
Correspondingly, in formula (2) and formula (3):
By above-mentioned formula (1)～(3) it can be seen that in the associating to early stage reverb signal
When the variance of probability density distribution is updated, the filter coefficient of employing is the filtering of t
Device coefficient (i.e. gfOr the g in formula (2)f(t)), at the connection completing reverb signal in early days
After the renewal of the variance closing probability density distribution, then the coefficient of wave filter is updated (i.e.
G in formula (2)f(t+1))。
Step 240, determines the audio signal after solving reverberation based on the filter coefficient after updating.
Use formula as above (1)～(3) joint probability density to early stage reverb signal
After the variance of distribution and the filter coefficient of sub-filter are updated, solve the sound after reverberation
Frequently signal dt,fEqual to when the spectral coefficient of the channel audio signal that t receives deducts t+1
The transposition of filter coefficient and the input signal of described sub-filter of the sub-filter carved to
The product of amount.
Exemplarily, audio signal d after reverberation is solvedT, fCan meet:
Additionally, in some optional implementations, the audio signal solution reverberation side of the present embodiment
Method can also comprise the following steps that
Judge whether the error between actual output and the desired output of sub-filter meets error
Absolute square more than the spectral coefficient of the channel audio signal received in t of absolute value
Value square with the product of predetermined threshold value K.In other words, it is judged that the reality of sub-filter is defeated
Go out whether the error e (t) between desired output meets | e (t) |2＞ K × | xt,f|2；
The most then filter coefficient is set to null vector.
Wherein, K is the threshold value pre-set, and meets K ＞ 1.
Additionally, work as | e (t) |2＞ K × | xt,f|2Time, it is believed that sub-filter dissipates, at next step
When the variance being distributed the joint probability density of early stage reverb signal is updated, formula (1)
gfFor null vector.
Using the audio signal solution reverberation method of the present embodiment, it solves amount of calculation needed for reverberation process
Less, and calculating process only needs between t-D～the t-L+1 moment channel audio signal of input
And the channel audio signal of present frame input so that the reality solving reverberation process of the present embodiment
Time property is stronger.
With further reference to Fig. 3, it illustrates another embodiment of audio signal solution reverberation method
Flow process 300.The flow process 300 of this audio signal solution reverberation method.
The present embodiment can have relative with step 210 in embodiment illustrated in fig. 2～step 240
The step 310 answered～step 340.
Unlike embodiment illustrated in fig. 2, the present embodiment may further comprise:
Step 350, the early stage if judged result in step 320 is no, then before updating
The wave filter of the sub-filter before the variance of the joint probability density distribution of reverb signal and renewal
Coefficient as update after early stage reverb signal joint probability density distribution variance and update after
The filter coefficient of sub-filter.
In other words, if step 320 is judged the single channel audio got in step 310
Signal is not voice signal, does not the most update the side of the joint probability density distribution of reverb signal in early days
Difference and the filter coefficient of sub-filter, and connection based on the early stage reverb signal before updating
The filter coefficient of the variance and sub-filter of closing probability density distribution obtains in step 310
The channel audio signal got carries out solving reverberation and processes (step 340).
Compared with the embodiment shown in Fig. 2, the audio signal solution reverberation method of the present embodiment is permissible
Avoid the output abnormality that in step 320, the error of judged result may cause, promote warp further
The late reverberation filter effect of the audio signal after the method process of the present embodiment.
With further reference to Fig. 4, as to the realization of method shown in above-mentioned each figure, the application provides
A kind of embodiment of audio signal solution reverberation unit, this device embodiment with shown in Fig. 2
Embodiment of the method corresponding, this device specifically can apply in various electronic equipment.
As shown in Figure 4, the audio signal solution reverberation unit 400 described in the present embodiment includes obtaining
Module 410, judge module 420, first more new module 430 and determine module 440:
Wherein:
Acquisition module 410 is configurable to obtain channel audio signal, channel audio signal
Including reverb signal and late reverberation signal in early days.
Judge module 420 is configurable to judge whether channel audio signal is voice signal.
If it is voice signal that the first more new module 430 is configurable to channel audio signal, then
Update the variance of the joint probability density distribution of reverb signal in early days, and based on early stage reverb signal
Joint probability density distribution variance update sub-filter filter coefficient, wherein, son
Band filter is for filtering the late reverberation signal comprised in channel audio signal.
After determining that module 440 is configurable to determine solution reverberation based on the filter coefficient after updating
Audio signal.
In some optional implementations, it is judged that module 420 can be configured to further: logical
Cross voice activity detection technology and judge whether channel audio signal is voice signal.
In some optional implementations,
The variance of the joint probability density distribution of reverb signal can be the subband filter before updating in early days
The product of the transposed matrix of the filter coefficient of ripple device and the input signal vector of sub-filter with
Spectral coefficient x at the channel audio signal that t receivest,fDifference absolute value square.
Exemplarily, the variance of the joint probability density distribution of reverb signal in early days can meet:
Wherein, xt,fThe spectral coefficient of the channel audio signal for receiving in t,For more
The transposition of the filter coefficient of the sub-filter before new,Defeated for described sub-filter
Enter signal vector；
Filter coefficient g after renewalf(t+1) for the filter coefficient before updating and renewal variable quantity
Sum.Wherein, update variable quantity be the first renewal running parameter and second update running parameter it
Ratio.First updates the reality that running parameter is the iteration step length of sub-filter, sub-filter
Error between output and desired output and the product of the input signal vector of sub-filter.
Second transposition updating the input signal vector that running parameter is sub-filter and sub-filter
The product of input signal vector.Mistake between actual output and the desired output of sub-filter
The poor spectral coefficient equal to the channel audio signal received in t and the early stage reverberation of t
The ratio of the variance of the joint probability density distribution of signal deducts the filtering of the sub-filter before renewal
The transposition of device coefficient and the product of the input signal vector of sub-filter.
Exemplarily, the filter coefficient g after renewalf(t+1) can meet:
Wherein, μ is the iteration step length of sub-filter, and e (t) is the actual output of sub-filter
And the error e (t) between desired output meets:
In some optional implementations, solve audio signal d after reverberationt,fEqual in t
The spectral coefficient of the channel audio signal received deducts the filtering of the sub-filter in t+1 moment
The product of the input signal vector of the transposition of device coefficient and described sub-filter.
Exemplarily, audio signal d after reverberation is solvedt,fCan meet:
In some optional implementations, the audio signal solution reverberation unit of the present embodiment also may be used
To farther include the second more new module (not shown).
If it is not voice signal that the second more new module is configurable to channel audio signal, then will
Subband filter before the variance of the joint probability density distribution of the early stage reverb signal before renewal and renewal
The filter coefficient of ripple device as update after early stage reverb signal joint probability density distribution
The filter coefficient of the sub-filter after variance and renewal.
In some optional implementations, the audio signal solution reverberation unit of the present embodiment also may be used
To farther include zero setting module (not shown).
Zero setting module is configurable to judge between actual output and the desired output of sub-filter
Error whether meet Error Absolute Value square more than the single channel audio that receives in t
The absolute value of the spectral coefficient of signal square with the product of predetermined threshold value K.In other words, it is judged that
Whether the error e (t) between actual output and the desired output of sub-filter meets | e (t) |2＞ K
×|xt,f|2；And the most then filter coefficient is set to null vector；Wherein, K is for set in advance
The threshold value put, and meet K ＞ 1.
It will be understood by those skilled in the art that above-mentioned audio signal solution reverberation unit 400 is all right
Including some other known features, such as processor, memorizer etc., in order to unnecessarily obscure
Embodiment of the disclosure, structure known to these is the most not shown.
Below with reference to Fig. 5, it illustrates the terminal unit be suitable to for realizing the embodiment of the present application
Or the structural representation of the computer system 500 of server.
As it is shown in figure 5, computer system 500 includes CPU (CPU) 501, its
Can be according to the program being stored in read only memory (ROM) 502 or from storage part 508
It is loaded into the program in random access storage device (RAM) 503 and performs various suitable action
And process.In RAM 503, also storage has system 500 to operate required various program sums
According to.CPU 501, ROM 502 and RAM 503 are connected with each other by bus 504.Input
/ output (I/O) interface 505 is also connected to bus 504.
It is connected to I/O interface 505: include the importation 506 of keyboard, mouse etc. with lower component；
Including such as cathode ray tube (CRT), liquid crystal display (LCD) etc. and speaker etc.
Output part 507；Storage part 508 including hard disk etc.；And include such as LAN card,
The communications portion 509 of the NIC of modem etc..Communications portion 509 is via such as
The network of the Internet performs communication process.Driver 510 is connected to I/O interface also according to needs
505.Detachable media 511, such as disk, CD, magneto-optic disk, semiconductor memory etc.,
Be arranged on as required in driver 510, in order to the computer program read from it according to
Needs are mounted into storage part 508.
Especially, according to embodiment of the disclosure, the process described above with reference to flow chart is permissible
It is implemented as computer software programs.Such as, embodiment of the disclosure and include a kind of computer journey
Sequence product, it includes the computer program being tangibly embodied on machine readable media, described meter
Calculation machine program comprises the program code for performing the method shown in flow chart.In such enforcement
In example, this computer program can be downloaded and installed from network by communications portion 509,
And/or be mounted from detachable media 511.
Flow chart in accompanying drawing and block diagram, it is illustrated that according to the various embodiment of the application system,
Architectural framework in the cards, function and the operation of method and computer program product.This point
On, each square frame in flow chart or block diagram can represent a module, program segment or code
A part, a part for described module, program segment or code comprise one or more for
Realize the executable instruction of the logic function of regulation.It should also be noted that at some as replacement
In realization, the function marked in square frame can also be sent out to be different from the order marked in accompanying drawing
Raw.Such as, two square frames succeedingly represented can essentially perform substantially in parallel, they
Sometimes can also perform in the opposite order, this is depending on involved function.It is also noted that
It is, the square frame in each square frame in block diagram and/or flow chart and block diagram and/or flow chart
Combination, can realize by the special hardware based system of the function or operation that perform regulation,
Or can realize with the combination of specialized hardware with computer instruction.
Being described in the embodiment of the present application involved module can be real by the way of software
Existing, it is also possible to realize by the way of hardware.Described module can also be arranged on process
In device, for example, it is possible to be described as: a kind of processor include acquisition module, judge module, the
One more new module and determine module.Wherein, the title of these modules the most not structure
The restriction of this module itself in pairs, such as, acquisition module is also described as " obtaining single-pass
The module of audio channel signal ".
As on the other hand, present invention also provides a kind of nonvolatile computer storage media,
This nonvolatile computer storage media can be described in above-described embodiment included in device
Nonvolatile computer storage media；Can also be individualism, be unkitted allocate in terminal non-
Volatile computer storage medium.Above-mentioned nonvolatile computer storage media storage have one or
The multiple program of person, when one or more program is performed by an equipment so that equipment: obtain
Taking channel audio signal, channel audio signal includes reverb signal and late reverberation letter in early days
Number；Judge whether channel audio signal is voice signal；The most then update the letter of reverberation in early days
Number the variance of joint probability density distribution, and joint probability density based on early stage reverb signal
The variance of distribution updates the filter coefficient of sub-filter, and wherein, sub-filter is used for filtering
Except the late reverberation signal comprised in channel audio signal；And based on the wave filter after updating
Coefficient determines the channel audio signal solving reverberation.
Above description is only the preferred embodiment of the application and saying institute's application technology principle
Bright.It will be appreciated by those skilled in the art that invention scope involved in the application, do not limit
In the technical scheme of the particular combination of above-mentioned technical characteristic, also should contain simultaneously without departing from
In the case of described inventive concept, above-mentioned technical characteristic or its equivalent feature carry out combination in any
And other technical scheme formed.Such as features described above and (but not limited to) disclosed herein
The technical characteristic with similar functions is replaced mutually and the technical scheme that formed.
Claims (12)
1. an audio signal solution reverberation method, it is characterised in that including:
Obtain channel audio signal, described channel audio signal include in early days reverb signal and
Late reverberation signal；
Judge whether described channel audio signal is voice signal；
The most then update the variance of the joint probability density distribution of described early stage reverb signal, and
The variance of joint probability density based on described early stage reverb signal distribution updates sub-filter
Filter coefficient, wherein, described sub-filter is used for filtering in described channel audio signal
The late reverberation signal comprised；And
The channel audio signal solving reverberation is determined based on the described filter coefficient after updating.
Method the most according to claim 1, it is characterised in that the described single-pass of described judgement
Whether audio channel signal is that voice signal includes:
Judge whether described channel audio signal is that voice is believed by voice activity detection technology
Number.
Method the most according to claim 1, it is characterised in that:
The variance of the joint probability density distribution of described early stage reverb signal is the described son before updating
The transposed matrix of the filter coefficient of band filter and the input signal vector of described sub-filter
The spectral coefficient x of product and the channel audio signal received in tt,fThe absolute value of difference
Square；
Filter coefficient g after renewalf(t+1) for the filter coefficient before updating and renewal variable quantity
Sum；
Wherein, described renewal variable quantity is that the first renewal running parameter and second updates running parameter
Ratio；
Described first renewal running parameter is the iteration step length of described sub-filter, described subband
Error between actual output and the desired output of wave filter and described sub-filter defeated
Enter the product of signal vector；
Described second transposition updating the input signal vector that running parameter is described sub-filter
Product with the input signal vector of described sub-filter；
Error between actual output and the desired output of described sub-filter is equal in t
The spectral coefficient of the channel audio signal received is general with combining of the early stage reverb signal of t
The ratio of the variance of rate Density Distribution deducts the filter coefficient of the described sub-filter before renewal
The product of the input signal vector of transposition and described sub-filter.
Method the most according to claim 1, it is characterised in that the sound after described solution reverberation
Frequently signal dt,fEqual to when the spectral coefficient of the channel audio signal that t receives deducts t+1
The transposition of filter coefficient and the input signal of described sub-filter of the sub-filter carved to
The product of amount.
5. according to the method described in claim 1-4 any one, it is characterised in that described
After judging whether described channel audio signal is voice signal, described method also includes:
If it is not, the side that the joint probability density of the described early stage reverb signal before then updating is distributed
The filter coefficient of the sub-filter before difference and renewal is believed as the early stage reverberation after described renewal
Number the variance of joint probability density distribution and the filter coefficient of sub-filter after updating.
6. according to the method described in claim 1-4 any one, it is characterised in that described side
Method also includes:
Judge whether the error between actual output and the desired output of described sub-filter meets
Error Absolute Value square more than the spectral coefficient of the channel audio signal that receives in t
Absolute value square with the product of predetermined threshold value K；
The most described filter coefficient is set to null vector；
Wherein, K ＞ 1.
7. an audio signal solution reverberation unit, it is characterised in that including:
Acquisition module, is configured to obtain channel audio signal, described channel audio signal
Including reverb signal and late reverberation signal in early days；
Judge module, is configured to judge whether described channel audio signal is voice signal；
First more new module, if being configured to described channel audio signal is voice signal, then
Update the variance of the joint probability density distribution of described early stage reverb signal, and based on described early stage
The variance of the joint probability density distribution of reverb signal updates the filter coefficient of sub-filter,
Wherein, described sub-filter mixes for filtering the late period comprised in described channel audio signal
Ring signal；And
Determine module, be configured to determine solution reverberation based on the described filter coefficient after updating
Channel audio signal.
Device the most according to claim 7, it is characterised in that described judge module enters one
Step is configured to:
Judge whether described channel audio signal is that voice is believed by voice activity detection technology
Number.
Device the most according to claim 7, it is characterised in that:
The variance of the joint probability density distribution of described early stage reverb signal is:
The transposed matrix of the filter coefficient of the described sub-filter before renewal is filtered with described subband
The spectrum of the product of the input signal vector of ripple device and the channel audio signal received in t
Coefficient xt,fDifference absolute value square；
Filter coefficient g after renewalf(t+1) for the filter coefficient before updating and renewal variable quantity
Sum；
Wherein, described renewal variable quantity is that the first renewal running parameter and second updates running parameter
Ratio；
Described first renewal running parameter is the iteration step length of described sub-filter, described subband
Error between actual output and the desired output of wave filter and described sub-filter defeated
Enter the product of signal vector；
Described second transposition updating the input signal vector that running parameter is described sub-filter
Product with the input signal vector of described sub-filter；
Error between actual output and the desired output of described sub-filter is equal in t
The spectral coefficient of the channel audio signal received is general with combining of the early stage reverb signal of t
The ratio of the variance of rate Density Distribution deducts the filter coefficient of the described sub-filter before renewal
The product of the input signal vector of transposition and described sub-filter.
Device the most according to claim 7, it is characterised in that after described solution reverberation
Audio signal dt,fT+1 is deducted equal to the spectral coefficient of the channel audio signal received in t
The transposition of the filter coefficient of the sub-filter in moment and the input signal of described sub-filter
The product of vector.
11. according to the device described in claim 7-10 any one, it is characterised in that described
Device also includes the second more new module；
If it is not voice letter that the described second more new module is configured to described channel audio signal
Number, then the variance and more of the joint probability density distribution of the described early stage reverb signal before updating
The filter coefficient of the sub-filter before new is as the connection of the early stage reverb signal after described renewal
The filter coefficient of the sub-filter after the variance of conjunction probability density distribution and renewal.
12. according to the device described in claim 7-10 any one, it is characterised in that described
Device also includes zero setting module；
It is defeated with expectation that described zero setting module is configured to the actual output judging described sub-filter
Error between going out whether meet Error Absolute Value square more than the single-pass that receives in t
The absolute value of the spectral coefficient of audio channel signal square with the product of predetermined threshold value K；And
The most described filter coefficient is set to null vector；
Wherein, K ＞ 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610474006.1A CN105931648B (en) | 2016-06-24 | 2016-06-24 | Audio signal solution reverberation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610474006.1A CN105931648B (en) | 2016-06-24 | 2016-06-24 | Audio signal solution reverberation method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105931648A true CN105931648A (en) | 2016-09-07 |
| CN105931648B CN105931648B (en) | 2019-05-03 |
Family
ID=56829221
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610474006.1A Active CN105931648B (en) | 2016-06-24 | 2016-06-24 | Audio signal solution reverberation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105931648B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107328410A (en) * | 2017-06-30 | 2017-11-07 | 百度在线网络技术（北京）有限公司 | Method and automobile computer for positioning automatic driving vehicle |
| CN107328411A (en) * | 2017-06-30 | 2017-11-07 | 百度在线网络技术（北京）有限公司 | Vehicle positioning system and automatic driving vehicle |
| CN111489760A (en) * | 2020-04-01 | 2020-08-04 | 腾讯科技（深圳）有限公司 | Speech signal dereverberation processing method, speech signal dereverberation processing device, computer equipment and storage medium |
| CN113223543A (en) * | 2021-06-10 | 2021-08-06 | 北京小米移动软件有限公司 | Speech enhancement method, apparatus and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2009535674A (en) * | 2006-05-01 | 2009-10-01 | 日本電信電話株式会社 | Method and apparatus for speech dereverberation based on stochastic model of sound source and room acoustics |
| CN103033815A (en) * | 2012-12-19 | 2013-04-10 | 中国科学院声学研究所 | Detection Method and detection device of distance expansion target based on reverberation covariance matrix |
| US20130297298A1 (en) * | 2012-05-04 | 2013-11-07 | Sony Computer Entertainment Inc. | Source separation using independent component analysis with mixed multi-variate probability density function |
| CN104995676A (en) * | 2013-02-14 | 2015-10-21 | 杜比实验室特许公司 | Signal decorrelation in an audio processing system |
| US20160005413A1 (en) * | 2013-02-14 | 2016-01-07 | Dolby Laboratories Licensing Corporation | Audio Signal Enhancement Using Estimated Spatial Parameters |
| WO2016014254A1 (en) * | 2014-07-23 | 2016-01-28 | Pcms Holdings, Inc. | System and method for determining audio context in augmented-reality applications |
-
2016
- 2016-06-24 CN CN201610474006.1A patent/CN105931648B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2009535674A (en) * | 2006-05-01 | 2009-10-01 | 日本電信電話株式会社 | Method and apparatus for speech dereverberation based on stochastic model of sound source and room acoustics |
| US20130297298A1 (en) * | 2012-05-04 | 2013-11-07 | Sony Computer Entertainment Inc. | Source separation using independent component analysis with mixed multi-variate probability density function |
| CN103033815A (en) * | 2012-12-19 | 2013-04-10 | 中国科学院声学研究所 | Detection Method and detection device of distance expansion target based on reverberation covariance matrix |
| CN104995676A (en) * | 2013-02-14 | 2015-10-21 | 杜比实验室特许公司 | Signal decorrelation in an audio processing system |
| US20160005413A1 (en) * | 2013-02-14 | 2016-01-07 | Dolby Laboratories Licensing Corporation | Audio Signal Enhancement Using Estimated Spatial Parameters |
| WO2016014254A1 (en) * | 2014-07-23 | 2016-01-28 | Pcms Holdings, Inc. | System and method for determining audio context in augmented-reality applications |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107328410A (en) * | 2017-06-30 | 2017-11-07 | 百度在线网络技术（北京）有限公司 | Method and automobile computer for positioning automatic driving vehicle |
| CN107328411A (en) * | 2017-06-30 | 2017-11-07 | 百度在线网络技术（北京）有限公司 | Vehicle positioning system and automatic driving vehicle |
| CN111489760A (en) * | 2020-04-01 | 2020-08-04 | 腾讯科技（深圳）有限公司 | Speech signal dereverberation processing method, speech signal dereverberation processing device, computer equipment and storage medium |
| CN111489760B (en) * | 2020-04-01 | 2023-05-16 | 腾讯科技（深圳）有限公司 | Speech signal dereverberation processing method, device, computer equipment and storage medium |
| CN113223543A (en) * | 2021-06-10 | 2021-08-06 | 北京小米移动软件有限公司 | Speech enhancement method, apparatus and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105931648B (en) | 2019-05-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107680586B (en) | Far-field speech acoustic model training method and system | |
| CN109155006A (en) | The audio analysis based on frequency is carried out using neural network | |
| CN105931648A (en) | Audio signal de-reverberation method and device | |
| CN102870156B (en) | Audio communication device, method for outputting an audio signal, and communication system | |
| CN108108902A (en) | A kind of risk case alarm method and device | |
| CN110428842A (en) | Speech model training method, device, equipment and computer readable storage medium | |
| CN106409310A (en) | Audio signal classification method and device | |
| CN103220440A (en) | Noise suppressing device, noise suppressing method, and program | |
| CN108877783A (en) | The method and apparatus for determining the audio types of audio data | |
| CN107277178A (en) | Method and apparatus for pushed information | |
| CN109242043A (en) | Method and apparatus for generating information prediction model | |
| CN114492831A (en) | Method and device for generating federal learning model | |
| CN107451869A (en) | Method and apparatus for pushed information | |
| CN108446658A (en) | The method and apparatus of facial image for identification | |
| CN109584895B (en) | Voice de-noising method and device | |
| CN113793620B (en) | Voice noise reduction method, device and equipment based on scene classification and storage medium | |
| CN106169961A (en) | The network parameter processing method and processing device of neutral net based on artificial intelligence | |
| CN109087124A (en) | A kind of application program Value Prediction Methods and device | |
| CN108062423A (en) | Information-pushing method and device | |
| CN109165570A (en) | Method and apparatus for generating information | |
| CN106340310B (en) | Speech detection method and device | |
| CN116485527A (en) | Financial credit risk analysis method and device | |
| Catellier et al. | Wenets: A convolutional framework for evaluating audio waveforms | |
| CN106341564A (en) | Method and device for processing signal data and intelligent conference equipment | |
| CN106488554A (en) | A kind of fingerprint database method for building up and system |
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 |