CN108053834A - audio data processing method, device, terminal and system - Google Patents
audio data processing method, device, terminal and system Download PDFInfo
- Publication number
- CN108053834A CN108053834A CN201711272872.3A CN201711272872A CN108053834A CN 108053834 A CN108053834 A CN 108053834A CN 201711272872 A CN201711272872 A CN 201711272872A CN 108053834 A CN108053834 A CN 108053834A
- Authority
- CN
- China
- Prior art keywords
- filtering
- data
- wiener
- voice
- processing
- 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
- 238000003672 processing method Methods 0.000 title claims abstract description 29
- 238000001914 filtration Methods 0.000 claims abstract description 509
- 238000012545 processing Methods 0.000 claims abstract description 178
- 238000000034 method Methods 0.000 claims abstract description 45
- 230000005764 inhibitory process Effects 0.000 claims abstract description 34
- 230000001052 transient effect Effects 0.000 claims description 52
- 238000001228 spectrum Methods 0.000 claims description 22
- 230000005540 biological transmission Effects 0.000 claims description 14
- 238000009499 grossing Methods 0.000 claims description 14
- 238000011946 reduction process Methods 0.000 claims description 12
- XVZUMQAMCYSUMS-SIUGBPQLSA-N OC(=O)C[C@@H](C(O)=O)NC(=O)[C@H](C)NC(=O)[C@H](C(C)C)NC(=O)[C@@H](N)CC1=CC=C(O)C=C1 Chemical compound OC(=O)C[C@@H](C(O)=O)NC(=O)[C@H](C)NC(=O)[C@H](C(C)C)NC(=O)[C@@H](N)CC1=CC=C(O)C=C1 XVZUMQAMCYSUMS-SIUGBPQLSA-N 0.000 claims description 9
- 241001269238 Data Species 0.000 claims description 8
- 230000006835 compression Effects 0.000 claims description 4
- 238000007906 compression Methods 0.000 claims description 4
- 235000013399 edible fruits Nutrition 0.000 claims 2
- 230000000694 effects Effects 0.000 abstract description 38
- 238000001514 detection method Methods 0.000 abstract description 30
- 230000004044 response Effects 0.000 abstract description 21
- 230000002452 interceptive effect Effects 0.000 abstract description 12
- 230000008569 process Effects 0.000 description 13
- 238000010586 diagram Methods 0.000 description 11
- 230000009467 reduction Effects 0.000 description 10
- 230000008901 benefit Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 5
- 230000003993 interaction Effects 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 238000001728 nano-filtration Methods 0.000 description 3
- 230000009471 action Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 239000004568 cement Substances 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000009021 linear effect Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 230000009022 nonlinear effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000009466 transformation 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
-
- 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
- G10L21/0232—Processing in the frequency domain
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Circuit For Audible Band Transducer (AREA)
Abstract
The embodiment of the invention discloses a kind of audio data processing method, device, terminal and system, this method includes：Obtain the voice data after space filtering；First Wiener filtering and the second Wiener filtering are carried out respectively to voice data, respectively obtain the first filtering data and the second filtering data, the first Wiener filtering is more than the second Wiener filtering to the inhibition level of noise；Sentenced using the first filtering data and stop the second filtering data, stop result to the progress data processing of the second filtering data according to sentencing.The embodiment of the present invention is according to voice activity detection and the different demands of automatic speech recognition, different degrees of Wiener filtering is carried out respectively, it can not only ensure the accuracy rate of automatic speech recognition, it can also avoid disturbing the influence to voice activity detection, more accurately detect Voice Activity Status, shorten the delay of feedback of interactive voice, improve the response speed to phonetic order, better usage experience is brought to user.
Description
Technical field
The present invention relates to technical field of data processing more particularly to a kind of audio data processing method, device, terminal and it is
System.
Background technology
Intelligent sound interaction is the important branch of artificial intelligence field, accurately intelligent sound interaction is realized freely, by pole
The both hands of big liberation people, obtain and the more free information flow of physical world and manipulation.Intelligent sound interacts main near field language
Sound interacts and far field interactive voice.Past 20 years, near field voice obtained tremendous expansion, and the phonetic recognization rate near field approaches at present
In the phonetic recognization rate of people, but more free interaction should be far field interactive voice.So-called far field interactive voice, it is meant that say
There is a certain distance between words people and interactive device, expand the free space of speaker, but this can introduce the excessive back of the body again
Scape noise jamming causes the intractability of voice activity detection and automatic speech recognition to be significantly greatly increased.
Voice activity detection detects the voice that speaker is actually typing from one section of continuous voice data.Accurately
Voice activity detection on the one hand can improve the accuracy rate of follow-up automatic speech recognition, on the other hand can also reduce voice friendship
Mutual delay of feedback, user speech instruction one terminate that implementing result can be provided rapidly, are brought to user and preferably use body
It tests.
At present, it is general original audio data is handled using array signal after, using treated voice data into
Row voice activity detection and automatic speech recognition, but should treated voice data still has certain interference, can serious shadow
The accuracy of voice activity detection is rung, causes the error of voice activity detection, so as to cause the response to phonetic order slow.
The content of the invention
In view of this, the present invention provides a kind of audio data processing method, device, terminal and system, can solve existing
Voice activity detection error caused by the error that the voice data that has in technology that treated still has, influences phonetic order sound
The problem of answering speed.
A kind of audio data processing method provided in an embodiment of the present invention, including：
Obtain the voice data after space filtering；
First Wiener filtering and the second Wiener filtering are carried out respectively to the voice data, respectively obtain the first filtering data
With the second filtering data, first Wiener filtering is more than second Wiener filtering to the inhibition level of noise；
Sentenced using first filtering data and stop second filtering data, stop result to the described second filtering number according to sentencing
According to progress data processing.
Optionally, it is described that first Wiener filtering and the second Wiener filtering are carried out respectively to the voice data, it respectively obtains
First filtering data and the second filtering data, specifically include：
Using the M powers of strength factor, first Wiener filtering is carried out to the voice data, obtains first filter
Wave number evidence；Using the Nth power of the strength factor, second Wiener filtering is carried out to the voice data, obtains described the
Two filtering datas；M is more than N.
Optionally, the M powers using strength factor carry out first Wiener filtering to the voice data, obtain
To first filtering data, specifically include：
According to formulaDescribed first is carried out to the voice data Y (j ω)
Wiener filtering obtains the second filtering data YVAD(jω)；
The Nth power using the strength factor carries out second Wiener filtering to the voice data, obtains institute
The second filtering data is stated, is specifically included：
According to formulaDescribed second is carried out to the voice data Y (j ω)
Wiener filtering obtains the second filtering data YASR(jω)；
Wherein, M=1, N=1/2, the strength factor areThe Pyy(j ω) is the voice data
Power spectrum, Pxx(j ω) is the average power spectra of the original audio data before the voice data space filtering, and EPS is minimum
Value.
Optionally, the utilization first filtering data, which is sentenced, stops second filtering data, further includes before：
Interference is carried out to first filtering data to handle；
It is described that interference is gone to handle, including transient noise Processing for removing, noise reduction process and noise smoothing processing in one or
It is multiple.
Optionally, the transient noise Processing for removing, specifically includes：
Obtain the voice data each increasing of corresponding first Wiener filtering of frequency domain point in default frequency domain
Benefit；
The quantity of voice data frequency domain point in the default frequency domain is counted, obtains the first value；Statistical gain
The quantity of frequency domain point of the amplitude within preset gain threshold value, obtains second value；
According to the described first value and the second value, obtain transient state and eliminate gain；
Gain is eliminated according to the transient state, eliminates the transient noise in first filtering data.
Optionally, the voice data obtained after space filtering, specifically includes：
Obtain the original audio data of sound pick-up outfit acquisition；
After carrying out Short Time Fourier Transform to the original audio data, obtain each passage in the sound pick-up outfit and correspond to
Frequency-region signal；
Space filtering processing is carried out to the corresponding frequency-region signal of each passage, obtains the audio after the space filtering
Data；
The utilization first filtering data, which is sentenced, stops second filtering data, specifically includes：
The inversion process of Short Time Fourier Transform is carried out to first filtering data and second filtering data；
Using treated the first filtering data sentences second filtering data that stops that treated, stop result to the place according to sentencing
The second filtering data after reason carries out data processing.
The embodiment of the present invention additionally provides a kind of audio data processing method, applied to first terminal equipment, the method,
Including：
Obtain the voice data after space filtering；
First Wiener filtering and the second Wiener filtering are carried out respectively to the voice data, respectively obtain the first filtering data
With the second filtering data, first Wiener filtering is more than second Wiener filtering to the inhibition level of noise；
First filtering data and second filtering data are sent to second terminal equipment, so that described second is whole
End equipment is sentenced using first filtering data stops second filtering data, and stops result to the described second filtering number according to sentencing
According to progress data processing.
Optionally, it is described that first Wiener filtering and the second Wiener filtering are carried out respectively to the voice data, it respectively obtains
First filtering data and the second filtering data, specifically include：
Using the M powers of strength factor, first Wiener filtering is carried out to the voice data, obtains first filter
Wave number evidence；Using the Nth power of the strength factor, second Wiener filtering is carried out to the voice data, obtains described the
Two filtering datas；M is more than N.
Optionally, the M powers using strength factor carry out first Wiener filtering to the voice data, obtain
To first filtering data, specifically include：
According to formulaFirst dimension is carried out to the voice data Y (j ω)
Nanofiltration ripple obtains the second filtering data YVAD(jω)；
The Nth power using the strength factor carries out second Wiener filtering to the voice data, obtains institute
The second filtering data is stated, is specifically included：
According to formulaDescribed second is carried out to the voice data Y (j ω)
Wiener filtering obtains the second filtering data YASR(jω)；
Wherein, M=1, N=1/2, the strength factor areThe Pyy(j ω) is the voice data
Power spectrum, Pxx(j ω) is the average power spectra of the original audio data before the voice data space filtering, and EPS is minimum
Value.
Optionally, the utilization first filtering data, which is sentenced, stops second filtering data, further includes before：
Interference is carried out to first filtering data to handle；
It is described that interference is gone to handle, including transient noise Processing for removing, noise reduction process and noise smoothing processing in one or
It is multiple.
Optionally, the transient noise Processing for removing, specifically includes：
Obtain the voice data each increasing of corresponding first Wiener filtering of frequency domain point in default frequency domain
Benefit；
The quantity of voice data frequency domain point in the default frequency domain is counted, obtains the first value；Statistical gain
The quantity of frequency domain point of the amplitude within the first preset gain threshold value, obtains second value；
According to the described first value and the second value, obtain transient state and eliminate gain；
Gain is eliminated according to the transient state, eliminates the transient noise in first filtering data.
Optionally, the voice data obtained after space filtering, specifically includes：
Obtain the original audio data of sound pick-up outfit acquisition；
After carrying out Short Time Fourier Transform to the original audio data, obtain each passage in the sound pick-up outfit and correspond to
Frequency-region signal；
Space filtering processing is carried out to the corresponding frequency-region signal of each passage, obtains the audio after the space filtering
Data.
Optionally, it is described that first filtering data is sent to second terminal equipment, it specifically includes：
The second terminal equipment is sent to after carrying out down-sampled processing to first filtering data.
Optionally, it is described that first filtering data and second filtering data are sent to second terminal equipment, tool
Body includes：
After first filtering data and second filtering data are packaged compression, it is sent to the second terminal and sets
It is standby.
A kind of audio-frequency data processing device provided in an embodiment of the present invention, including：Data acquisition module, the first filtering mould
Block, the second filter module and first processing module；
The data acquisition module, for obtaining the voice data after space filtering；
First filter module for carrying out the first Wiener filtering to the voice data, obtains the first filtering data；
Second filter module for carrying out the second Wiener filtering to the voice data, obtains the second filtering data,
First Wiener filtering is more than second Wiener filtering to the inhibition level of noise；
The first processing module stops second filtering data for sentencing using first filtering data, according to sentencing
Stop result and data processing is carried out to second filtering data.
Optionally, first filter module, is specifically used for：
Using the M powers of strength factor, first Wiener filtering is carried out to the voice data, obtains first filter
Wave number evidence；
Second filter module, is specifically used for：
Using the Nth power of the strength factor, second Wiener filtering is carried out to the voice data, obtains described the
Two filtering datas；
Wherein, M is more than N.
Optionally, first filter module, including：First processing submodule；
The first processing submodule, for according to formulaTo the audio
Data Y (j ω) carries out first Wiener filtering, obtains the second filtering data YVAD(jω)；
Second filter module, including：Second processing submodule；
The second processing submodule, for according to formulaTo the audio
Data Y (j ω) carries out second Wiener filtering, obtains the second filtering data YASR(jω)；
Wherein, M=1, N=1/2, the strength factor areThe Pyy(j ω) is the audio number
According to power spectrum, Pxx(j ω) is the average power spectra of the original audio data before the voice data space filtering, and EPS is pole
Small value.
Optionally, described device further includes：Second processing module；
The Second processing module is handled for carrying out interference to first filtering data；It is described that interference is gone to handle,
Including one or more of transient noise Processing for removing, noise reduction process and noise smoothing processing；
The first processing module, specifically for treated that the first filtering data sentences stops using the Second processing module
Second filtering data stops result to second filtering data progress audio identification according to sentencing.
Optionally, when it is described go interference processing include the transient noise Processing for removing when, the Second processing module, have
Body includes：Noise reduction submodule；
The noise reduction submodule, is specifically used for：
Obtain the voice data each increasing of corresponding first Wiener filtering of frequency domain point in default frequency domain
Benefit；The quantity of voice data frequency domain point in the default frequency domain is counted, obtains the first value；Statistical gain amplitude exists
The quantity of frequency domain point within first preset gain threshold value, obtains second value；According to the described first value and the second value, obtain
Transient state eliminates gain；Gain is eliminated according to the transient state, eliminates the transient noise in first filtering data.
Optionally, the data acquisition module, is specifically used for：
Obtain the original audio data of sound pick-up outfit acquisition；Short Time Fourier Transform is carried out to the original audio data
Afterwards, the corresponding frequency-region signal of each passage in the sound pick-up outfit is obtained；The corresponding frequency-region signal of each passage is carried out
Space filtering processing, obtains the voice data after the space filtering；
The first processing module, is specifically used for：
The inversion process of Short Time Fourier Transform is carried out to first filtering data and second filtering data；Profit
With treated the first filtering data sentences second filtering data that stops that treated, stop result according to sentencing treated second to described
Filtering data carries out audio identification data processing.
The embodiment of the present invention additionally provides a kind of audio-frequency data processing device, applied to first terminal equipment, including：Data
Acquisition module, the first filter module, the second filter module and data transmission module；
The data acquisition module, for obtaining the voice data after space filtering；
First filter module for carrying out the first Wiener filtering to the voice data, obtains the first filtering data；
Second filter module for carrying out the second Wiener filtering to the voice data, obtains the second filtering data,
First Wiener filtering is more than second Wiener filtering to the inhibition level of noise；
The data transmission module, for first filtering data and second filtering data to be sent to second eventually
End equipment, so that the first filtering data described in the second terminal equipment utilization, which is sentenced, stops second filtering data, and according to sentencing
Stop result and data processing is carried out to second filtering data.
Optionally, first filter module, is specifically used for：
Using the M powers of strength factor, first Wiener filtering is carried out to the voice data, obtains first filter
Wave number evidence；
Second filter module, is specifically used for：
Using the Nth power of the strength factor, second Wiener filtering is carried out to the voice data, obtains described the
Two filtering datas；
Wherein, M is more than N.
Optionally, first filter module, including：First processing submodule；
The first processing submodule, for according to formulaTo the audio
Data Y (j ω) carries out first Wiener filtering, obtains the second filtering data YVAD(jω)；
Second filter module, including：Second processing submodule；
The second processing submodule, for according to formulaTo the audio
Data Y (j ω) carries out second Wiener filtering, obtains the second filtering data YASR(jω)；
Wherein, M=1, N=1/2, the strength factor areThe Pyy(j ω) is the voice data
Power spectrum, Pxx(j ω) is the average power spectra of the original audio data before the voice data space filtering, and EPS is minimum
Value.
Optionally, described device further includes：Data processing module；
The data processing module is handled for carrying out interference to first filtering data；It is described that interference is gone to handle,
Including one or more of transient noise Processing for removing, noise reduction process and noise smoothing processing；
The data transmission module, specifically for by the data processing module treated the first filtering data and described
Second filtering data is sent to second terminal equipment so that treated described in the second terminal equipment utilization first filtering number
Stop second filtering data according to sentencing, and stop result to second filtering data progress data processing according to sentencing.
Optionally, when it is described go interference processing include the transient noise Processing for removing when, the data processing module, have
Body includes：Noise reduction submodule；
The noise reduction submodule, is specifically used for：
Obtain the voice data each increasing of corresponding first Wiener filtering of frequency domain point in default frequency domain
Benefit；The quantity of voice data frequency domain point in the default frequency domain is counted, obtains the first value；Statistical gain amplitude exists
The quantity of frequency domain point within first preset gain threshold value, obtains second value；According to the described first value and the second value, obtain
Transient state eliminates gain；Gain is eliminated according to the transient state, eliminates the transient noise in first filtering data.
Optionally, the data acquisition module, is specifically used for：
Obtain the original audio data of sound pick-up outfit acquisition；Short Time Fourier Transform is carried out to the original audio data
Afterwards, the corresponding frequency-region signal of each passage in the sound pick-up outfit is obtained；The corresponding frequency-region signal of each passage is carried out
Space filtering processing, obtains the voice data after the space filtering.
Optionally, which is characterized in that the data transmission module is specifically used for：
The second terminal equipment is sent to after carrying out down-sampled processing to first filtering data.
Optionally, data transmission module, also particularly useful for：
After first filtering data and second filtering data are packaged compression, it is sent to the second terminal and sets
It is standby.
The embodiment of the present invention additionally provides a kind of voice data processing terminal, including：Memory and processor；
Said program code for storing program code, and is transmitted to the processor by the memory；
The processor for the instruction in said program code, performs as claimed in any one of claims 1 to 6
Audio data processing method.
The embodiment of the present invention additionally provides a kind of audio-frequency data processing system, including：First equipment and the second equipment；
First equipment, for obtaining the voice data after space filtering；Be additionally operable to the voice data respectively into
The first Wiener filtering of row and the second Wiener filtering obtain the first filtering data and the second filtering data, first Wiener filtering
Second Wiener filtering is more than to the inhibition level of noise；It is additionally operable to first filtering data and the second filtering number
According to being sent to the second equipment；
Second equipment is stopped second filtering data for being sentenced using first filtering data, and stopped according to sentencing
As a result audio identification is carried out to second filtering data.
Compared with prior art, the present invention has at least the following advantages：
In embodiments of the present invention, carry out the Wiener filtering of varying strength respectively to the voice data after space filtering, obtain
The filtering data different to noise suppressed degree to two-way, i.e., the first filtering data more to noise suppressed and to noise
Smaller second filtering data of inhibition level.Then, sentenced using the first filtering data and stop the second filtering data, according to judging result
Data processing is carried out to the second filtering data.It, can be larger due to higher for the inhibition level of noise in the first filtering data
Degree avoids influence of the interference to voice activity detection, improves the precision of voice activity detection and the response of automatic speech recognition
Speed.And the inhibition level in the second filtering data for noise is relatively low, it can be to avoid higher noise suppressed to speech recognition
The influence of accuracy.The embodiment of the present invention carries out not respectively according to voice activity detection and the different demands of automatic speech recognition
With the Wiener filtering of degree, it can not only ensure the accuracy rate of automatic speech recognition, interference can also be avoided to examine speech activity
The influence of survey, more accurately detects Voice Activity Status, shortens the delay of feedback of interactive voice, improves to phonetic order
Response speed brings better usage experience to user.
Description of the drawings
It in order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, below will be to embodiment or existing
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments described in application, for those of ordinary skill in the art, without creative efforts,
It can also be obtained according to these attached drawings other attached drawings.
Fig. 1 is a kind of schematic diagram of existing treated voice data；
Fig. 2 is a kind of flow diagram of audio data processing method provided in an embodiment of the present invention；
Fig. 3 is the signal of original audio data, the first filtering data and the second filtering data in the specific embodiment of the invention
Figure；
Fig. 4 is the flow diagram of another audio data processing method provided in an embodiment of the present invention；
Fig. 5 provides the flow diagram of middle transient noise Processing for removing for the specific embodiment of the invention；
Fig. 6 is the flow diagram of another audio data processing method provided in an embodiment of the present invention；
Fig. 7 is a kind of structure diagram of audio-frequency data processing device provided in an embodiment of the present invention；
Fig. 8 is the structure diagram of another audio-frequency data processing device provided in an embodiment of the present invention.
Specific embodiment
In order to which those skilled in the art is made to more fully understand the present invention program, below in conjunction in the embodiment of the present invention
The technical solution in the embodiment of the present invention is clearly and completely described in attached drawing, it is clear that described embodiment is only this
Invention part of the embodiment, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art exist
All other embodiments obtained under the premise of creative work are not made, belong to the scope of protection of the invention.
In order to make it easy to understand, multiple technical terms involved in the embodiment of the present invention are introduced first.
Voice activity detection (Voice Activity Detection, VAD) is also known as speech terminals detection, the inspection of voice border
It surveys, refers to the presence or absence that target voice is detected in noise circumstance.Commonly used in voice coding, speech enhan-cement, speech recognition
It waits in speech processing systems.
Automatic speech recognition (Automatic Speech Recognition, ASR) is a kind of to be converted to the voice of people
The technology of text.
It is usually that voice data all the way is directly exported using array signal processing, both during current interactive voice
Speech recognition is carried out as ASR data, and the beginning and end (sentence and stop) of ASR processing is judged as VAD data.However, by
Also the accuracy of VAD processing is influenced, for details, reference can be made to Fig. 1, this is illustrated there are more interference in the voice data of output
A kind of existing treated voice data.It can be seen from figure 1 that the phonetic order that user actually enters originate in node A,
End at node B.And carry out the obtained phonetic order of VAD processing using the data and end at node B ', i.e., at node B '
It can just judge phonetic order end of data, then carry out ASR processing according to the data between node A and node B '.This not only can
Cause it is slow to the response speed of phonetic order, can also influence ASR processing speed, cause the response to phonetic order slow.
The present inventor has found that ASR processing needs to ensure that the voice data of input keeps smaller as far as possible under study for action
Non-linear distortion, and VAD processing needs to carry out interference the inhibition of higher degree, but higher to the inhibition of interference,
It can then cause the non-linear distortion introduced more.However, if treated, data are higher to the inhibition level of noise, can be because non-
Linear effect causes voice data to be distorted, and reduces the accuracy of speech recognition；If ensureing the accuracy of speech recognition, count
Sentencing for VAD can be influenced there are more serious interference again in stop as a result, cause VAD detection errors, influencing the sound of phonetic order
Answer speed.That is, same input data cannot meet the need of ASR processing and VAD processing to input data noise suppressed degree simultaneously
It asks.
For this purpose, an embodiment of the present invention provides a kind of audio data processing method, device, terminal and system, according to ASD and
The different noise suppressed demand of VAD processing, carries out the voice data after space filtering the Wiener filtering of varying strength, obtains respectively
To the two paths of data different to noise suppressed degree, VAD processing is carried out using to the higher filtered data of noise suppressed degree,
Phonetic order sentence stopping, ASR processing is carried out using to the relatively low filtered data of noise suppressed degree, is ensureing at ASR
On the basis of managing accuracy rate, the signal component for being unfavorable for VAD processing in data is reduced to the greatest extent, improves VAD processing and is sentenced
The accuracy stopped, so as to improve the speed of ASR processing and the response speed to phonetic order.
Based on above-mentioned thought, in order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to
Attached drawing is described in detail the specific embodiment of the present invention.
Referring to Fig. 2, which is a kind of flow diagram of audio data processing method provided in an embodiment of the present invention.
Firstly the need of explanation, a kind of audio data processing method provided in an embodiment of the present invention can be applied to appoint
One terminal device, the terminal device can configure or are connected with multichannel radio reception microphone, to receive phonetic order input by user.
As an example, which can be specifically smart mobile phone, tablet computer, PC and server etc., here not
It enumerates again.
Audio data processing method provided in an embodiment of the present invention specifically includes following steps S201-S203.
S201：Obtain the voice data Y (j ω) after space filtering.
Since in practical applications, voice (such as phonetic order) input by user is generally by multiple microphone radio reception
It obtains, is disturbed in the original audio data that radio reception obtains in different spaces and microphone channel different.Therefore, in order to correspond to sky
Between it is different to remove the noise jamming in original audio data, it is necessary to according to the different to carrying out space filtering processing of space,
With the voice data (the voice data Y (j ω) i.e. after space filtering) after interference of tentatively being gone out.
It should also be noted that, before space filtering is carried out, first voice input by user can also be pre-processed,
It can specifically include going direct current biasing, windowed function etc., then audio carries out Short Time Fourier Transform (Short to treated
Time Fourier Transform, STFT), obtain frequency domain signal X of the voice input by user under different frequency domains1(jω)、
X2(jω)、......、XN(jω).Then, to the frequency domain signal X1(jω)、X2(jω)、......、XN(j ω) carries out space
Filtering process can obtain Y (j ω).
That is, in the possible realization method of the embodiment of the present invention, step S201 can specifically include：Sound pick-up outfit is obtained to adopt
The original audio data of collection；After carrying out Short Time Fourier Transform to original audio data, each passage pair in sound pick-up outfit is obtained
The frequency domain signal X answered1(jω)、X2(jω)、......、XN(jω)；Space filter is carried out to the corresponding frequency-region signal of each passage
Ripple processing, obtains the voice data Y after space filtering (j ω).
In the specific implementation, can utilize general matching law (General Sidelobe Cancellation,
) or minimum variance distortionless response (Minimum Variance Distortionless Response, MVDR) wave beam shape GSC
It grows up to be a useful person to obtain Y (j ω), which is not described herein again for specific processing method.
S202：First Wiener filtering and the second Wiener filtering are carried out respectively to the voice data after space filtering, respectively
To the first filtering data and the second filtering data, the first Wiener filtering is more than the second Wiener filtering to the inhibition level of noise.
In embodiments of the present invention, the voice data noise suppressed degree handled with VAD to input is handled in order to meet ASR
Different demands, for ASR processing and VAD processing, varying strength is carried out respectively to the voice data Y (j ω) after space filtering
Wiener filtering, with ensure ASR processing and VAD processing accuracy.To voice data Y (j ω) carry out noise suppressed degree compared with
The first high Wiener filtering reduces the signal component for being unfavorable for VAD processing in voice data Y (j ω) to the greatest extent, reduces the
To the interference of VAD processing in one filtering data, the accuracy that VAD is handled is improved.Noise suppressed is carried out to voice data Y (j ω)
The second relatively low Wiener filtering of degree avoids influence of the distortion to ASR processing, ensures the identification accuracy of ASR processing.In this way,
It can not only ensure the identification accuracy to phonetic order, the accuracy of VAD processing can also be improved, phonetic order is done in time
Go out response,
In practical applications, can by adjusting Wiener filtering strength factor with to the voice data after space filtering into
The Wiener filtering of the different noise suppressed degree of row, obtains the first filtering data and the second filtering data.
In some possible realization methods of the embodiment of the present invention, above-mentioned steps S202 can specifically include following step
Suddenly：
Using the M powers of strength factor, the first Wiener filtering is carried out to the voice data Y (j ω) after space filtering, is obtained
First filtering data；Using the Nth power of strength factor, the second wiener filter is carried out to the voice data Y (j ω) after space filtering
Ripple obtains the second filtering data；Wherein, M is more than N.
It should be noted that strength factor is influenced in Wiener filtering to the inhibition level of noise, strength factor is bigger to making an uproar
The inhibition level of sound is higher.Therefore, in embodiments of the present invention, the first dimension is carried out using the different power sides of same strength factor
Nanofiltration ripple and the second Wiener filtering, you can obtain first filtering data and second filtering data different to noise suppressed degree.
As an example, which can be set toM and N take 1 and 1/2 respectively.Wherein,
Pyy(j ω) is the power spectrum of voice data, can specifically be acquired by following formula (1)；Pxx(j ω) is voice data in space filtering
The average power spectra of preceding original audio data can specifically be acquired by following formula (2)；EPS is minimum.
Pyy(j ω)=α Pyy(jω)+(1-α)Y(jω)Y*(jω) (1)
In practical applications, firm power spectrum P can be obtained using single order smooth mannerxx(j ω) and Pyy(jω)。
Then, the first Wiener filtering is carried out to voice data Y (j ω), obtains the second filtering data YVAD(j ω), specifically can be with
It utilizes following formula (3)：
The second Wiener filtering then is carried out to voice data Y (j ω), obtains the second filtering data YASR(j ω), specifically can be with
It utilizes following formula (4)：
S203：Sentenced using the first filtering data and stop the second filtering data, stop result to the progress of the second filtering data according to sentencing
Data processing.
It is understood that eliminating the influence to VAD processing in the first filtering data, the first filtering data is utilized
To target voice (such as phonetic order) is accurately identified in the second filtering data, sentence and stop the second filtering data.According to sentencing
The result stopped carries out data processing (such as automatic speech recognition) to the second filtering data, can ensure the base of processing accuracy
On plinth, the response speed of data processing is improved.
It should also be noted that, in the specific implementation, need that frequency-region signal carries out by treated before step S203
The inverse transformation (Inverse Short-Time Fourier Transform, ISTFT) of Short Time Fourier Transform obtains time domain letter
Number, it is superimposed to obtain the first filtering data and the second filtering data by adding window, then sentence and stop and data processing.
With reference to concrete scene, the above-mentioned advantage for embodiment that the present invention will be described in detail.Referring to Fig. 3, the figure shows this
Original audio data, the first filtering data and the second filtering data in invention specific embodiment.Wherein, original audio data carries
Phonetic order and interference obtain the first filtering after the first higher Wiener filtering of noise suppressed intensity is carried out to original audio data
Data obtain the second filtering data after the second relatively low Wiener filtering of noise suppressed intensity is carried out to original audio data.From Fig. 3
In can be seen that, can ensure the accuracy of speech recognition using the second filtering data；It can be improved and sentenced using the first Wiener filtering
The accuracy stopped, more accurately detects Voice Activity Status, shortens the delay of feedback of interactive voice, improves to phonetic order
Response speed, bring better usage experience to user.
In embodiments of the present invention, carry out the Wiener filtering of varying strength respectively to the voice data after space filtering, obtain
The filtering data different to noise suppressed degree to two-way, i.e., the first filtering data more to noise suppressed and to noise
Smaller second filtering data of inhibition level.Then, sentenced using the first filtering data and stop the second filtering data, according to judging result
Data processing is carried out to the second filtering data.It, can be larger due to higher for the inhibition level of noise in the first filtering data
Degree avoids influence of the interference to voice activity detection, improves the precision of voice activity detection and the response of automatic speech recognition
Speed.And the inhibition level in the second filtering data for noise is relatively low, it can be to avoid higher noise suppressed to speech recognition
The influence of accuracy.The embodiment of the present invention carries out not respectively according to voice activity detection and the different demands of automatic speech recognition
With the Wiener filtering of degree, it can not only ensure the accuracy rate of automatic speech recognition, interference can also be avoided to examine speech activity
The influence of survey, more accurately detects Voice Activity Status, shortens the delay of feedback of interactive voice, improves to phonetic order
Response speed brings better usage experience to user.
Referring to Fig. 4, which is the flow diagram of another audio data processing method provided in an embodiment of the present invention.
In order to further improve the accuracy sentenced and stopped, the embodiment of the present invention further includes before step S203：
S204：Interference is carried out to the first filtering data to handle.
In embodiments of the present invention, it is above-mentioned that interference is gone to handle, can specifically it include：At transient noise Processing for removing, noise reduction
One or more of reason and noise smoothing processing.When it is implemented, transient noise can be performed one by one to the first filtering data
Processing for removing, noise reduction process and noise smoothing processing.
The processing of transient noise Processing for removing, noise reduction process and noise smoothing is illustrated below.
First, transient noise Processing for removing, as shown in figure 5, specifically comprising the following steps S501-S503.
S501：Obtain voice data each gain of corresponding first Wiener filtering of frequency domain point in default frequency domain.
In the embodiment of the present application, the gain of the first Wiener filtering, that is, above-mentioned strength factor corresponds to the value with each frequency domain.
As an example, the gain of corresponding first Wiener filtering of each frequency domain point isCorresponding to each frequency domain
Value.
S502：The quantity of voice data frequency domain point in default frequency domain is counted, obtains the first value；Statistical gain amplitude
The quantity of frequency domain point within preset gain threshold value, obtains second value.
S503：According to the first value and second value, obtain transient state and eliminate gain.
It should be noted that since high frequency attenuation and reflection cause high frequency randomness higher, it is steady to obtain higher
Property, the ratio that gain within certain frequency is less than some threshold value is only counted, noise is carried out to the first filtering data on this basis
Smoothing processing.
When it is implemented, 0-2000Hz can be taken by presetting flat frequency domain, preset gain threshold value can take 0.3.
In one example, transient state eliminates gain gain and can be obtained according to following formula (5).
Wherein, all_bin is the first value, and count_bin is second value.
S504：Gain is eliminated according to transient state, eliminates the transient noise in the first filtering data.
In embodiments of the present invention, the first filter can be eliminated by the way that transient state elimination gain will be applied to the first filtering data
Transient noise of the wave number in.
Secondly, any one noise reduction algorithm specifically may be employed in noise reduction process, no longer repeats one by one here.
Finally, noise smoothing is handled, and can specifically be realized by carrying out noise estimation to the first filtering data.
In one example, its single order smooth power spectrum first is calculated to the first filtering data of every frame after windowing process
Pnoise(j ω) is specifically acquired using above formula (1).Then, the single order smooth power spectrum of more each the first filtering data of frame,
More new historical minimum power composes minPnoise(j ω), such as following formula (6),
Wherein, β and ρ is coefficient.
The single order smooth power spectrum of the frame is estimated as to the noise of several frames of the starting of the first filtering data (such as 50 frames)
Pnoise(jω)；Current history minimum power spectrum min P are estimated as to the noise of each frame after several framesnoise(jω).And
Afterwards, the noise that the frame is superimposed on each frame of the first filtering data is estimated, you can the noise for making the first filtering data is steady, keeps away
Exempt from the error that noise mutation causes VAD to handle.
Based on the audio data processing method that above-described embodiment provides, the embodiment of the present invention additionally provides another audio number
According to processing method, by the responsible original audio data of first terminal equipment (such as smart mobile phone, tablet computer, server etc.)
Reason, obtains the first filtering data and the second filtering data, is responsible for sentencing by second terminal equipment (such as server) and stop and data processing
Process can not only ensure that the data operation quantity in first terminal equipment will not be excessive, can also be adopted in second terminal equipment
With more responsible VAD Processing Algorithms, more accurate VAD handling results are obtained.
Specifically, referring to Fig. 6, which is that the flow of another audio data processing method provided in an embodiment of the present invention is shown
It is intended to.
A kind of audio data processing method provided in an embodiment of the present invention applied to first terminal equipment, can specifically wrap
Include following steps S601-S603.
S601：Obtain the voice data after space filtering.
Optionally, step S601 is specifically included：Obtain the original audio data of sound pick-up outfit acquisition；To original audio number
After Short Time Fourier Transform is carried out, the corresponding frequency-region signal of each passage in sound pick-up outfit is obtained；It is corresponding to each passage
Frequency-region signal carries out space filtering processing, obtains the voice data after space filtering.
S602：First Wiener filtering and the second Wiener filtering are carried out respectively to voice data, respectively obtain the first filtering number
According to the second filtering data, the first Wiener filtering is more than the second Wiener filtering to the inhibition level of noise.
In the possible realization method of the embodiment of the present invention, step S602 can specifically include：
Using the M powers of strength factor, the first Wiener filtering is carried out to voice data, obtains the first filtering data；It utilizes
The Nth power of strength factor carries out the second Wiener filtering to voice data, obtains the second filtering data；Wherein, M is more than N.
Optionally, using the M powers of strength factor, the first Wiener filtering is carried out to voice data, obtains the first filtering number
According to specifically including：
According to formulaFirst Wiener filtering is carried out to voice data Y (j ω),
Obtain the second filtering data YVAD(jω)。
Using the Nth power of strength factor, the second Wiener filtering is carried out to voice data, obtains the second filtering data, specifically
Including：
According to formulaSecond Wiener filtering is carried out to voice data Y (j ω),
Obtain the second filtering data YASR(jω)。
Wherein, M=1, N=1/2, strength factor arePyy(j ω) be voice data power spectrum, Pxx
(j ω) is the average power spectra of the original audio data before voice data space filtering, and EPS is minimum.
It is understood that step S601-S602 and the step S201-S205 classes in above-described embodiment in the present embodiment
Seemingly, referring specifically to above-mentioned related description, repeat no more.
In the possible realization method of the embodiment of the present invention, further included before step S603：First filtering data is carried out
Interference is gone to handle.Specifically, interference is gone to handle, transient noise Processing for removing, noise reduction process and noise smoothing processing can be included
One or more of.
Wherein, as an example, transient noise Processing for removing can specifically include：
Obtain voice data each gain of corresponding first Wiener filtering of frequency domain point in default frequency domain；Count sound
Frequency obtains the first value according to the quantity of the frequency domain point in default frequency domain；Statistical gain amplitude is in the first preset gain threshold value
Within frequency domain point quantity, obtain second value；According to the first value and second value, obtain transient state and eliminate gain；Disappear according to transient state
Except gain, the transient noise in the first filtering data is eliminated.
It is understood that interference processing is gone with going interference processing class described in above-described embodiment in the present embodiment
Seemingly, referring specifically to related description, which is not described herein again.
S603：First filtering data and the second filtering data are sent to second terminal equipment, so that second terminal equipment
Sentenced using the first filtering data and stop the second filtering data, and stop result to the progress data processing of the second filtering data according to sentencing.
It is understood that filtering process to voice data and, using filtered data sentence and stop and data
Processing is responsible for execution by different terminal devices (or server) respectively, can not only ensure accuracy and the place of filtering process
Speed is managed, more complicated vad algorithm and ASR Algorithm can also be used, ensures to obtain and accurately sentences the speech recognition for stopping result
As a result.
In the possible realization method of the embodiment of the present invention, in order to reduce the transmission speed that the transmission quantity of data improves data
Degree, and then the response speed to phonetic order is improved, the first filtering data is uploaded to server, can specifically be included：
Second terminal equipment is sent to after carrying out down-sampled processing to the first filtering data.
As an example, the sample rate of the first filtering data can be reduced to 8kHz by 16kHz.
Optionally, the first filtering data and the second filtering data are uploaded to server, specifically included：
After first filtering data and the second filtering data are packaged compression, second terminal equipment is sent to.
In embodiments of the present invention, first terminal equipment carries out varying strength respectively to the voice data after space filtering
Wiener filtering obtains the two-way filtering data different to noise suppressed degree, i.e., the first filtering more to noise suppressed
Data and to smaller second filtering data of noise suppressed degree, are sent to second terminal equipment.Then, second terminal equipment profit
Sentenced with the first filtering data and stop the second filtering data, data processing is carried out to the second filtering data according to judging result.Due to
One filtering data of inhibition level in to(for) noise is higher, can largely avoid shadow of the interference to voice activity detection
It rings, improves the precision of voice activity detection and the response speed of automatic speech recognition.And for noise in the second filtering data
Inhibition level is relatively low, can be to avoid influence of the higher noise suppressed to speech discrimination accuracy.At the filtering of voice data
Reason and, using filtered data sentence and stop and data processing, born respectively by first terminal equipment and second terminal equipment
Duty performs, and can not only ensure the accuracy and processing speed of filtering process, can also use more complicated vad algorithm and
ASR Algorithm ensures to obtain and accurately sentences the voice recognition result for stopping result.The embodiment of the present invention is according to voice activity detection and certainly
The different demands of dynamic speech recognition, carry out different degrees of Wiener filtering, can not only ensure the standard of automatic speech recognition respectively
True rate can also avoid disturbing the influence to voice activity detection, more accurately detect Voice Activity Status, shorten voice
Interactive delay of feedback improves the response speed to phonetic order, and better usage experience is brought to user.
Based on the audio data processing method that above-described embodiment provides, the embodiment of the present invention additionally provides a kind of voice data
Processing unit.
Referring to Fig. 7, which is a kind of structure diagram of audio-frequency data processing device provided in an embodiment of the present invention.
A kind of audio-frequency data processing device provided in an embodiment of the present invention, including：Data acquisition module 701, first filters
Module 702, the second filter module 703 and first processing module 704；
Data acquisition module 701, for obtaining the voice data after space filtering；
First filter module 702, the voice data for being obtained to data acquisition module 701 carry out the first Wiener filtering,
Obtain the first filtering data；
Second filter module 703, the voice data for being obtained to data acquisition module 701 carry out the second Wiener filtering,
The second filtering data is obtained, the first Wiener filtering is more than the second Wiener filtering to the inhibition level of noise；
First processing module 704 stops the second filtering data for sentencing using the first filtering data, stops result to according to sentencing
Two filtering datas carry out data processing.
In the possible realization method of the embodiment of the present invention, the first filter module 702 is specifically used for：Utilize strength factor
M powers, to voice data carry out the first Wiener filtering, obtain the first filtering data；
Second filter module 703, is specifically used for：Using the Nth power of strength factor, the second wiener filter is carried out to voice data
Ripple obtains the second filtering data；Wherein, M is more than N.
In the possible realization method of the embodiment of the present invention, the first filter module 702, including：First processing submodule；
First processing submodule, for according to formulaTo voice data Y (j
The first Wiener filtering ω) is carried out, obtains the second filtering data YVAD(jω)；
Second filter module 703, including：Second processing submodule；
Second processing submodule, for according to formulaTo voice data Y (j
The second Wiener filtering ω) is carried out, obtains the second filtering data YASR(jω)；
Wherein, M=1, N=1/2, strength factor arePyy(j ω) be voice data power spectrum, Pxx
(j ω) is the average power spectra of the original audio data before voice data space filtering, and EPS is minimum.
In the possible realization method of the embodiment of the present invention, which further includes：Second processing mould
Block；
Second processing module is handled for carrying out interference to the first filtering data；Interference is gone to handle, including transient noise
One or more of Processing for removing, noise reduction process and noise smoothing processing；
First processing module 704, specifically for treated that the first filtering data sentences stops second using Second processing module
Filtering data stops result to the second filtering data progress audio identification according to sentencing.
Optionally, when interference processing is gone to include transient noise Processing for removing, Second processing module specifically includes：Noise reduction
Submodule；
Noise reduction submodule, is specifically used for：
Obtain voice data each gain of corresponding first Wiener filtering of frequency domain point in default frequency domain；Count sound
Frequency obtains the first value according to the quantity of the frequency domain point in default frequency domain；Statistical gain amplitude is in the first preset gain threshold value
Within frequency domain point quantity, obtain second value；According to the first value and second value, obtain transient state and eliminate gain；Disappear according to transient state
Except gain, the transient noise in the first filtering data is eliminated.
In the possible realization method of the embodiment of the present invention, data acquisition module 701 is specifically used for：
Obtain the original audio data of sound pick-up outfit acquisition；After carrying out Short Time Fourier Transform to original audio data, obtain
The corresponding frequency-region signal of each passage into sound pick-up outfit；Space filtering processing is carried out to the corresponding frequency-region signal of each passage,
Obtain the voice data after space filtering；
First processing module 704, is specifically used for：
The inversion process of Short Time Fourier Transform is carried out to the first filtering data and the second filtering data；After processing
The first filtering data sentence second filtering data that stops that treated, stop result according to sentencing the second filtering data carry out to treated
Audio identification data processing.
In embodiments of the present invention, carry out the Wiener filtering of varying strength respectively to the voice data after space filtering, obtain
The filtering data different to noise suppressed degree to two-way, i.e., the first filtering data more to noise suppressed and to noise
Smaller second filtering data of inhibition level.Then, sentenced using the first filtering data and stop the second filtering data, according to judging result
Data processing is carried out to the second filtering data.It, can be larger due to higher for the inhibition level of noise in the first filtering data
Degree avoids influence of the interference to voice activity detection, improves the precision of voice activity detection and the response of automatic speech recognition
Speed.And the inhibition level in the second filtering data for noise is relatively low, it can be to avoid higher noise suppressed to speech recognition
The influence of accuracy.The embodiment of the present invention carries out not respectively according to voice activity detection and the different demands of automatic speech recognition
With the Wiener filtering of degree, it can not only ensure the accuracy rate of automatic speech recognition, interference can also be avoided to examine speech activity
The influence of survey, more accurately detects Voice Activity Status, shortens the delay of feedback of interactive voice, improves to phonetic order
Response speed brings better usage experience to user.
Based on audio data processing method and device that above-described embodiment provides, the embodiment of the present invention additionally provides another kind
Audio-frequency data processing device.
Referring to Fig. 8, which is the structure diagram of another audio-frequency data processing device provided in an embodiment of the present invention.
A kind of audio-frequency data processing device provided in an embodiment of the present invention, applied to first terminal equipment, including：Data obtain
Modulus block 801, the first filter module 802, the second filter module 803 and data transmission module 804；
Data acquisition module 801, for obtaining the voice data after space filtering；
First filter module 802, the voice data for being obtained to data acquisition module 801 carry out the first Wiener filtering,
Obtain the first filtering data；
Second filter module 803, the voice data for being obtained to data acquisition module 801 carry out the second Wiener filtering,
The second filtering data is obtained, the first Wiener filtering is more than the second Wiener filtering to the inhibition level of noise；
Data transmission module 804, for the first filtering data and the second filtering data to be sent to second terminal equipment, with
Sentence the first filtering data of second terminal equipment utilization and stop the second filtering data, and according to sentence stop result to the second filtering data into
Row data processing.
In the possible realization method of the embodiment of the present invention, the first filter module 802 is specifically used for：Utilize strength factor
M powers, to voice data carry out the first Wiener filtering, obtain the first filtering data；Second filter module 803, is specifically used for：
Using the Nth power of strength factor, the second Wiener filtering is carried out to voice data, obtains the second filtering data；Wherein, M is more than N.
Optionally, the first filter module 802, including：First processing submodule；Second filter module 801, including：Second
Handle submodule；
First processing submodule, for according to formulaTo voice data Y (j
The first Wiener filtering ω) is carried out, obtains the second filtering data YVAD(jω)；
Second processing submodule, for according to formulaTo voice data Y (j
The second Wiener filtering ω) is carried out, obtains the second filtering data YASR(jω)；
Wherein, M=1, N=1/2, strength factor arePyy(j ω) be voice data power spectrum, Pxx
(j ω) is the average power spectra of the original audio data before voice data space filtering, and EPS is minimum.
In the possible realization method of the embodiment of the present invention, which further includes：Data processing mould
Block；
Data processing module is handled for carrying out interference to the first filtering data；Interference is gone to handle, including transient noise
One or more of Processing for removing, noise reduction process and noise smoothing processing；
Data transmission module 804, specifically for by data processing module treated the first filtering data and the second filtering
Data sending is to second terminal equipment, so that treated that the first filtering data sentences stops the second filtering number for second terminal equipment utilization
According to, and stop result to the progress data processing of the second filtering data according to sentencing.
Optionally, when interference processing is gone to include transient noise Processing for removing, data processing module specifically includes：Noise reduction
Submodule；
Noise reduction submodule, is specifically used for：
Obtain voice data each gain of corresponding first Wiener filtering of frequency domain point in default frequency domain；Count sound
Frequency obtains the first value according to the quantity of the frequency domain point in default frequency domain；Statistical gain amplitude is in the first preset gain threshold value
Within frequency domain point quantity, obtain second value；According to the first value and second value, obtain transient state and eliminate gain；Disappear according to transient state
Except gain, the transient noise in the first filtering data is eliminated.
In the possible realization method of the embodiment of the present invention, data acquisition module 801 is specifically used for：
Obtain the original audio data of sound pick-up outfit acquisition；After carrying out Short Time Fourier Transform to original audio data, obtain
The corresponding frequency-region signal of each passage into sound pick-up outfit；Space filtering processing is carried out to the corresponding frequency-region signal of each passage,
Obtain the voice data after space filtering.
In the possible realization method of the embodiment of the present invention, data transmission module 804 is specifically used for：To the first filtering number
According to being sent to second terminal equipment after carrying out down-sampled processing.
Optionally, data transmission module 804, also particularly useful for：First filtering data and the second filtering data are packaged pressure
After contracting, second terminal equipment is sent to.
In embodiments of the present invention, first terminal equipment carries out varying strength respectively to the voice data after space filtering
Wiener filtering obtains the two-way filtering data different to noise suppressed degree, i.e., the first filtering more to noise suppressed
Data and to smaller second filtering data of noise suppressed degree, are sent to second terminal equipment.Then, second terminal equipment profit
Sentenced with the first filtering data and stop the second filtering data, data processing is carried out to the second filtering data according to judging result.Due to
One filtering data of inhibition level in to(for) noise is higher, can largely avoid shadow of the interference to voice activity detection
It rings, improves the precision of voice activity detection and the response speed of automatic speech recognition.And for noise in the second filtering data
Inhibition level is relatively low, can be to avoid influence of the higher noise suppressed to speech discrimination accuracy.At the filtering of voice data
Reason and, using filtered data sentence and stop and data processing, born respectively by first terminal equipment and second terminal equipment
Duty performs, and can not only ensure the accuracy and processing speed of filtering process, can also use more complicated vad algorithm and
ASR Algorithm ensures to obtain and accurately sentences the voice recognition result for stopping result.The embodiment of the present invention is according to voice activity detection and certainly
The different demands of dynamic speech recognition, carry out different degrees of Wiener filtering, can not only ensure the standard of automatic speech recognition respectively
True rate can also avoid disturbing the influence to voice activity detection, more accurately detect Voice Activity Status, shorten voice
Interactive delay of feedback improves the response speed to phonetic order, and better usage experience is brought to user.
Based on audio data processing method and device that above-described embodiment provides, the embodiment of the present invention additionally provides a kind of sound
Frequency data processing terminal.The voice data processing terminal, including：Memory and processor.Wherein, memory, for storing journey
Sequence code, and program code is transmitted to processor；Processor for the instruction in program code, performs such as above-mentioned
Audio data processing method described in embodiment of anticipating.
Based on audio data processing method and device that above-described embodiment provides, the embodiment of the present invention additionally provides a kind of sound
Frequency data handling system.The audio-frequency data processing system, including：First equipment and the second equipment；
First equipment, for obtaining the voice data after space filtering；It is additionally operable to carry out the first dimension respectively to voice data
Nanofiltration ripple and the second Wiener filtering obtain the first filtering data and the second filtering data, inhibition of first Wiener filtering to noise
Degree is more than the second Wiener filtering；It is additionally operable to the first filtering data and the second filtering data being sent to the second equipment；
Second equipment stops the second filtering data for sentencing using the first filtering data, and stops result to the second filter according to sentencing
Wave number is according to progress audio identification.
It should be noted that each embodiment is described by the way of progressive in this specification, each embodiment emphasis is said
Bright is all difference from other examples, and just to refer each other for identical similar portion between each embodiment.For reality
For applying method, apparatus disclosed in example or system, since it is corresponded to the methods disclosed in the examples, so description is simpler
Single, reference may be made to the description of the method.
It should also be noted that, herein, relational terms such as first and second and the like are used merely to one
Entity or operation are distinguished with another entity or operation, without necessarily requiring or implying between these entities or operation
There are any actual relationship or orders.Moreover, term " comprising ", "comprising" or its any other variant are intended to contain
Lid non-exclusive inclusion, so that process, method, article or equipment including a series of elements not only will including those
Element, but also including other elements that are not explicitly listed or further include as this process, method, article or equipment
Intrinsic element.In the absence of more restrictions, the element limited by sentence " including one ... ", it is not excluded that
Also there are other identical elements in the process, method, article or apparatus that includes the element.
It can directly be held with reference to the step of method or algorithm that the embodiments described herein describes with hardware, processor
The combination of capable software module or the two is implemented.Software module can be placed in random access memory (RAM), memory, read-only deposit
Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technology
In any other form of storage medium well known in field.
The above described is only a preferred embodiment of the present invention, not make limitation in any form to the present invention.Though
So the present invention is disclosed above with preferred embodiment, however is not limited to the present invention.It is any to be familiar with those skilled in the art
Member, without departing from the scope of the technical proposal of the invention, all using the methods and technical content of the disclosure above to the present invention
Technical solution makes many possible changes and modifications or is revised as the equivalent embodiment of equivalent variations.Therefore, it is every without departing from
The content of technical solution of the present invention, technical spirit according to the invention is to any simple modification made for any of the above embodiments, equivalent
Variation and modification, still fall within technical solution of the present invention protection in the range of.
Claims (13)
1. a kind of audio data processing method, which is characterized in that the method, including：
Obtain the voice data after space filtering；
First Wiener filtering and the second Wiener filtering are carried out respectively to the voice data, respectively obtain the first filtering data and
Two filtering datas, first Wiener filtering are more than second Wiener filtering to the inhibition level of noise；
Sentenced using first filtering data and stop second filtering data, according to sentence stop result to second filtering data into
Row data processing.
2. according to the method described in claim 1, it is characterized in that, described carry out the voice data the first wiener filter respectively
Ripple and the second Wiener filtering, respectively obtain the first filtering data and the second filtering data, specifically include：
Using the M powers of strength factor, first Wiener filtering is carried out to the voice data, obtains the first filtering number
According to；Using the Nth power of the strength factor, second Wiener filtering is carried out to the voice data, obtains second filter
Wave number evidence；M is more than N.
3. according to the method described in claim 2, it is characterized in that,
The M powers using strength factor carry out first Wiener filtering to the voice data, obtain first filter
Wave number evidence, specifically includes：
According to formulaThe first wiener filter is carried out to the voice data Y (j ω)
Ripple obtains the second filtering data YVAD(jω)；
The Nth power using the strength factor carries out second Wiener filtering to the voice data, obtains described the
Two filtering datas, specifically include：
According to formulaSecond wiener is carried out to the voice data Y (j ω)
Filtering, obtains the second filtering data YASR(jω)；
Wherein, M=1, N=1/2, the strength factor areThe Pyy(j ω) is the work(of the voice data
Rate is composed, Pxx(j ω) is the average power spectra of the original audio data before the voice data space filtering, and EPS is minimum.
4. according to the method described in claim 1, it is characterized in that, the utilization first filtering data, which is sentenced, stops described second
Filtering data further includes before：
Interference is carried out to first filtering data to handle；
It is described that interference is gone to handle, including one or more of transient noise Processing for removing, noise reduction process and noise smoothing processing.
5. according to the method described in claim 4, it is characterized in that, the transient noise Processing for removing, specifically includes：
Obtain the voice data each gain of corresponding first Wiener filtering of frequency domain point in default frequency domain；
The quantity of voice data frequency domain point in the default frequency domain is counted, obtains the first value；Statistical gain amplitude
The quantity of frequency domain point within preset gain threshold value, obtains second value；
According to the described first value and the second value, obtain transient state and eliminate gain；
Gain is eliminated according to the transient state, eliminates the transient noise in first filtering data.
6. a kind of audio data processing method, which is characterized in that applied to first terminal equipment, the method, including：
Obtain the voice data after space filtering；
First Wiener filtering and the second Wiener filtering are carried out respectively to the voice data, respectively obtain the first filtering data and
Two filtering datas, first Wiener filtering are more than second Wiener filtering to the inhibition level of noise；
First filtering data and second filtering data are sent to second terminal equipment, so that the second terminal is set
It is standby sentenced using first filtering data stops second filtering data, and according to sentence stop result to second filtering data into
Row data processing.
7. according to the method described in claim 6, it is characterized in that, described carry out the voice data the first wiener filter respectively
Ripple and the second Wiener filtering, respectively obtain the first filtering data and the second filtering data, specifically include：
Using the M powers of strength factor, first Wiener filtering is carried out to the voice data, obtains the first filtering number
According to；Using the Nth power of the strength factor, second Wiener filtering is carried out to the voice data, obtains second filter
Wave number evidence；M is more than N.
8. according to the method described in claim 6, it is characterized in that, the utilization first filtering data, which is sentenced, stops described second
Filtering data further includes before：
Interference is carried out to first filtering data to handle；
It is described that interference is gone to handle, including one or more of transient noise Processing for removing, noise reduction process and noise smoothing processing.
9. according to the method described in claim 6-8 any one, which is characterized in that described by first filtering data and institute
It states the second filtering data and is sent to second terminal equipment, specifically include：
Down-sampled processing is carried out to first filtering data；
Treated the first filtering data and second filtering data are sent to the second terminal equipment；It alternatively, will place
After the first filtering data and second filtering data after reason are packaged compression, the second terminal equipment is sent to.
10. a kind of audio-frequency data processing device, which is characterized in that described device, including：Data acquisition module, the first filtering mould
Block, the second filter module and first processing module；
The data acquisition module, for obtaining the voice data after space filtering；
First filter module for carrying out the first Wiener filtering to the voice data, obtains the first filtering data；
Second filter module for carrying out the second Wiener filtering to the voice data, obtains the second filtering data, described
First Wiener filtering is more than second Wiener filtering to the inhibition level of noise；
The first processing module is stopped second filtering data for being sentenced using first filtering data, stops tying according to sentencing
Fruit carries out data processing to second filtering data.
11. a kind of audio-frequency data processing device, which is characterized in that applied to first terminal equipment, described device, including：Data
Acquisition module, the first filter module, the second filter module and data transmission module；
The data acquisition module, for obtaining the voice data after space filtering；
First filter module for carrying out the first Wiener filtering to the voice data, obtains the first filtering data；
Second filter module for carrying out the second Wiener filtering to the voice data, obtains the second filtering data, described
First Wiener filtering is more than second Wiener filtering to the inhibition level of noise；
The data transmission module, for first filtering data and second filtering data to be sent to second terminal and set
It is standby, so that the first filtering data described in the second terminal equipment utilization, which is sentenced, stops second filtering data, and stop tying according to sentencing
Fruit carries out data processing to second filtering data.
12. a kind of voice data processing terminal, which is characterized in that including：Memory and processor；
Said program code for storing program code, and is transmitted to the processor by the memory；
The processor for the instruction in said program code, performs sound as claimed in any one of claims 1 to 6
Frequency data processing method.
13. a kind of audio-frequency data processing system, which is characterized in that including：First equipment and the second equipment；
First equipment, for obtaining the voice data after space filtering；It is additionally operable to carry out the voice data respectively
One Wiener filtering and the second Wiener filtering obtain the first filtering data and the second filtering data, and first Wiener filtering is to making an uproar
The inhibition level of sound is more than second Wiener filtering；It is additionally operable to first filtering data and second filtering data hair
It send to the second equipment；
Second equipment is stopped second filtering data for being sentenced using first filtering data, and stops result according to sentencing
Audio identification is carried out to second filtering data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711272872.3A CN108053834B (en) | 2017-12-05 | 2017-12-05 | Audio data processing method, device, terminal and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711272872.3A CN108053834B (en) | 2017-12-05 | 2017-12-05 | Audio data processing method, device, terminal and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108053834A true CN108053834A (en) | 2018-05-18 |
| CN108053834B CN108053834B (en) | 2020-02-21 |
Family
ID=62121836
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711272872.3A Active CN108053834B (en) | 2017-12-05 | 2017-12-05 | Audio data processing method, device, terminal and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108053834B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111103568A (en) * | 2019-12-10 | 2020-05-05 | 北京声智科技有限公司 | Sound source positioning method, device, medium and equipment |
| CN111884729A (en) * | 2020-07-17 | 2020-11-03 | 上海动听网络科技有限公司 | Recording channel selection method and device and electronic equipment |
| CN113823311A (en) * | 2021-08-19 | 2021-12-21 | 安徽创变信息科技有限公司 | Voice recognition method and device based on audio enhancement |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1967659A (en) * | 2005-11-14 | 2007-05-23 | 北京大学科技开发部 | Speech enhancement method applied to deaf-aid |
| EP2096769A1 (en) * | 2008-02-26 | 2009-09-02 | Sony Corporation | Communication device, noise removing method, and program |
| JP2009540378A (en) * | 2006-06-14 | 2009-11-19 | シーメンス アウディオローギッシェ テヒニク ゲゼルシャフト ミット ベシュレンクテル ハフツング | Signal separator, method for determining an output signal based on a microphone signal, and computer program |
| CN103813251A (en) * | 2014-03-03 | 2014-05-21 | 深圳市微纳集成电路与系统应用研究院 | Hearing-aid denoising device and method allowable for adjusting denoising degree |
| CN104954595A (en) * | 2015-05-15 | 2015-09-30 | 百度在线网络技术（北京）有限公司 | Cancellation method and device of residual echo |
-
2017
- 2017-12-05 CN CN201711272872.3A patent/CN108053834B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1967659A (en) * | 2005-11-14 | 2007-05-23 | 北京大学科技开发部 | Speech enhancement method applied to deaf-aid |
| JP2009540378A (en) * | 2006-06-14 | 2009-11-19 | シーメンス アウディオローギッシェ テヒニク ゲゼルシャフト ミット ベシュレンクテル ハフツング | Signal separator, method for determining an output signal based on a microphone signal, and computer program |
| EP2096769A1 (en) * | 2008-02-26 | 2009-09-02 | Sony Corporation | Communication device, noise removing method, and program |
| CN103813251A (en) * | 2014-03-03 | 2014-05-21 | 深圳市微纳集成电路与系统应用研究院 | Hearing-aid denoising device and method allowable for adjusting denoising degree |
| CN104954595A (en) * | 2015-05-15 | 2015-09-30 | 百度在线网络技术（北京）有限公司 | Cancellation method and device of residual echo |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111103568A (en) * | 2019-12-10 | 2020-05-05 | 北京声智科技有限公司 | Sound source positioning method, device, medium and equipment |
| CN111884729A (en) * | 2020-07-17 | 2020-11-03 | 上海动听网络科技有限公司 | Recording channel selection method and device and electronic equipment |
| CN111884729B (en) * | 2020-07-17 | 2022-03-01 | 上海动听网络科技有限公司 | Recording channel selection method and device and electronic equipment |
| CN113823311A (en) * | 2021-08-19 | 2021-12-21 | 安徽创变信息科技有限公司 | Voice recognition method and device based on audio enhancement |
| CN113823311B (en) * | 2021-08-19 | 2023-11-21 | 广州市盛为电子有限公司 | Voice recognition method and device based on audio enhancement |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108053834B (en) | 2020-02-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10373609B2 (en) | Voice recognition method and apparatus | |
| CN111161752B (en) | Echo cancellation method and device | |
| KR102002681B1 (en) | Bandwidth extension based on generative adversarial networks | |
| CN110600017A (en) | Training method of voice processing model, voice recognition method, system and device | |
| CN109326299B (en) | Speech enhancement method, device and storage medium based on full convolution neural network | |
| WO2018223727A1 (en) | Voiceprint recognition method, apparatus and device, and medium | |
| CN105654955B (en) | Audio recognition method and device | |
| CN108053834A (en) | audio data processing method, device, terminal and system | |
| US20190385630A1 (en) | Deep neural network based speech enhancement | |
| US20190378529A1 (en) | Voice processing method, apparatus, device and storage medium | |
| CN111415653B (en) | Method and device for recognizing speech | |
| CN111968651A (en) | WT (WT) -based voiceprint recognition method and system | |
| KR102204975B1 (en) | Method and apparatus for speech recognition using deep neural network | |
| CN113299306A (en) | Echo cancellation method, echo cancellation device, electronic equipment and computer-readable storage medium | |
| WO2018154372A1 (en) | Sound identification utilizing periodic indications | |
| CN110797008A (en) | Far-field speech recognition method, speech recognition model training method and server | |
| US20230116052A1 (en) | Array geometry agnostic multi-channel personalized speech enhancement | |
| Lan et al. | Research on speech enhancement algorithm of multiresolution cochleagram based on skip connection deep neural network | |
| Goswami et al. | A novel approach for design of a speech enhancement system using auto-trained NLMS adaptive filter | |
| Kamarudin et al. | Acoustic echo cancellation using adaptive filtering algorithms for Quranic accents (Qiraat) identification | |
| CN111477248B (en) | Audio noise detection method and device | |
| CN114220430A (en) | Multi-sound-zone voice interaction method, device, equipment and storage medium | |
| Lu et al. | Temporal modulation normalization for robust speech feature extraction and recognition | |
| CN113380244A (en) | Intelligent adjustment method and system for playing volume of equipment | |
| Dai et al. | 2D Psychoacoustic modeling of equivalent masking for automatic speech recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |