CN101727911B - Noise suppression device and noise suppression method - Google Patents
Noise suppression device and noise suppression method Download PDFInfo
- Publication number
- CN101727911B CN101727911B CN2009102055351A CN200910205535A CN101727911B CN 101727911 B CN101727911 B CN 101727911B CN 2009102055351 A CN2009102055351 A CN 2009102055351A CN 200910205535 A CN200910205535 A CN 200910205535A CN 101727911 B CN101727911 B CN 101727911B
- Authority
- CN
- China
- Prior art keywords
- noise
- gain
- ratio
- input signal
- noise suppression
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
- 230000001629 suppression Effects 0.000 title claims abstract description 148
- 238000000034 method Methods 0.000 title claims abstract description 47
- 238000001228 spectrum Methods 0.000 claims abstract description 115
- 238000009499 grossing Methods 0.000 claims description 54
- 238000001514 detection method Methods 0.000 claims description 21
- LCIYFINKFGDAHD-UHFFFAOYSA-N azepane;3-nitrobenzoic acid Chemical compound C1CCCNCC1.OC(=O)C1=CC=CC([N+]([O-])=O)=C1 LCIYFINKFGDAHD-UHFFFAOYSA-N 0.000 description 68
- 230000000694 effects Effects 0.000 description 34
- 238000012545 processing Methods 0.000 description 33
- 230000005236 sound signal Effects 0.000 description 22
- 238000011282 treatment Methods 0.000 description 15
- 238000002203 pretreatment Methods 0.000 description 11
- 230000008569 process Effects 0.000 description 11
- 230000009471 action Effects 0.000 description 10
- 230000008859 change Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 7
- 230000000875 corresponding effect Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 230000002123 temporal effect Effects 0.000 description 6
- 230000009466 transformation Effects 0.000 description 6
- 230000004069 differentiation Effects 0.000 description 5
- 238000012937 correction Methods 0.000 description 4
- 230000006866 deterioration Effects 0.000 description 4
- 230000009467 reduction Effects 0.000 description 4
- 230000033228 biological regulation Effects 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000003595 spectral effect Effects 0.000 description 3
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 2
- 230000005534 acoustic noise Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 235000008733 Citrus aurantifolia Nutrition 0.000 description 1
- 235000012364 Peperomia pellucida Nutrition 0.000 description 1
- 240000007711 Peperomia pellucida Species 0.000 description 1
- 235000011941 Tilia x europaea Nutrition 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000009432 framing Methods 0.000 description 1
- 230000007274 generation of a signal involved in cell-cell signaling Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 230000002045 lasting effect Effects 0.000 description 1
- 239000004571 lime Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 239000002023 wood Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
- G10L21/0232—Processing in the frequency domain
-
- 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
- G10L19/04—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 predictive techniques
- G10L19/08—Determination or coding of the excitation function; Determination or coding of the long-term prediction parameters
- G10L19/083—Determination or coding of the excitation function; Determination or coding of the long-term prediction parameters the excitation function being an excitation gain
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0264—Noise filtering characterised by the type of parameter measurement, e.g. correlation techniques, zero crossing techniques or predictive techniques
Landscapes
- Engineering & Computer Science (AREA)
- Human Computer Interaction (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Quality & Reliability (AREA)
- Circuit For Audible Band Transducer (AREA)
- Noise Elimination (AREA)
- Soundproofing, Sound Blocking, And Sound Damping (AREA)
Abstract
The present invention provides a noise suppression device and a noise suppression method, wherein the noise suppression device does not generate music noise. Furthermore, even at a state of switching between an audio frame and a noise frame, the variation of the noise is natural. The noise suppression device comprises the following components: a noise spectrum calculation part (20) which aims at K frequency bands (wherein, K is a natural number which is larger than or equal to 2) for respectively calculating the noise spectrum contained in the input signal; a noise suppression gain calculation part (30) which is based on the calculation result for calculating the respective noise suppression gains of K frequency bands; and a primary sound adding unit (referring symbols 50*60 in the chart) which adds the input signal into a first noise suppression back signal with a first ratio. The first ratio is determined based on a difference between a noise suppression gain and a noise suppression gain. The first noise suppression back signal is obtained through applying the noise suppression gain to the input signal.
Description
Technical field
The present invention relates to a kind of Noise Suppression Device and noise suppressing method.
Background technology
Current, following apparatus etc. has been proposed, that is: audio playing apparatus, itself and input signal drive the loudspeaker even load accordingly; VOCA voice communications assembly, it is transmitting audio frequency away from intersite; And speech recognizing device, it is through distinguishing identification to kind of audio frequency etc., thereby understands its content etc.In above-mentioned each device, in order audio frequency to be play exactly, to be transmitted or identification etc., and preferably remove the The noise that is included in wherein.
As the technology of such squelch, known below for example patent documentation 1 and 2 and non-patent literature 1 to 5 in disclosed technology.
Patent documentation 1: TOHKEMY 2007-226264 communique
Patent documentation 2: TOHKEMY 2005-257748 communique
Non-patent literature 1:Boll, S., " Suppression of acoustic noise in speechusing spectral subtraction ", IEEE Trans.Vol.ASSP-27, No2, pp.113-120,1979.
Non-patent literature 2:M.Berouti, el al, " Enhancement of SpeechCorrupted by Acoustic Noise ", Proceedings of ICASSP, PP.201-211,1979.
Non-patent literature 3:Lim & Oppenheim, " Enhancement and BandwidthCompression of Noisy Speech ", Proc.IEEE, Vol67, No12, pp.1586-1604,1979
Non-patent literature 4:Y.Ephraim and D.Malah; " Speech EnhancementUsing a Minimum Mean-Square Error Short-Time Spectral AmplitudeEstimator ", IEEE Trans.Vol.ASSP-32, No.6; Pp1109-1121,1984.
Summary of the invention
Disclosed technology in above-mentioned each document; Basically relevant with following method; That is, the level that deducts the amplitude frequency spectrum of frequency band through appropriate method suppresses the method for noise, promptly so-called spectrum-subtraction (Spectral Subtraction); Each technology through in these can obtain certain noise suppression effect.
But, in these documents, still have unexposed unsolved problem.
For example, above-mentioned spectrum-subtraction carries out based on following manner,, infers the noise spectrum that is included in the input signal that is, from amplitude frequency spectrum, deducts this noise spectrum presumed value, but in the case, has the problem that produces so-called music noise probably.This is because said here noise spectrum presumed value not necessarily reflects actual noise spectrum.That is, exist in certain frequency band, deduct after the noise spectrum presumed value the still situation of residual noise, in addition, exist in other the frequency band, produce the excessive situation of cutting down.Therefore, if the amplitude frequency spectrum that will deduct behind the noise spectrum carries out conversion again to time domain, then might occur the synthetic signal of a plurality of sine waves with random frequency through it is play, possibly produced very ear-piercing noise (that is music noise).
In addition, proposed following technology, that is,, carried out adding in the amplitude frequency spectrum after deducting the noise spectrum presumed value processing (with reference to above-mentioned non-patent literature 5) of former sound in order to suppress this music noise.
But, definite because the additional ratio of former sound is based on the SNR that infers input signal in this technology, so exist squelch to handle the shortcoming that lacks some so-called dirigibilities.For example, in above-mentioned non-patent literature 5, owing to be to be that 10~15 [dB] are as benchmark with SNR; At SNR during more than or equal to this benchmark; Making the additional ratio of former sound is 0.05, and during less than this benchmark, making the additional ratio of former sound is such mode such as 0.5 grade at SNR; Irrespectively carry out the setting of the additional ratio of former sound with original expectation noisiness that suppress or that should suppress, so there are some problems in the noise suppression effect that in fact obtains.Above-mentioned situation also exists in above-mentioned patent documentation 2.
In addition, in input signal, there are mainly part (audio-frequency unit) that occupies by audio frequency and the part (noise section) that does not almost have audio frequency.
Under above-mentioned condition, for example, shown in above-mentioned non-patent literature 1; To audio-frequency unit spectrum of use subtraction, to the gain of noise section application of fixed, suppress thus under the situation of noise, when this current gain value is too small; From noise section under the situation that audio-frequency unit switches, produce the phenomenon that amount of background noise increases, and when current gain value is excessive; On the contrary, possibly produce the phenomenon that amount of background noise reduces.If it is for example play, then make the listener feel not sense naturally acoustically probably.
Problem of the present invention is that a kind of Noise Suppression Device and noise suppressing method that can solve at least a portion in the above-mentioned problem is provided.
In order to solve above-mentioned problem, Noise Suppression Device involved in the present invention has: noise spectrum is inferred the unit, and it infers the noise spectrum that is included in this input signal based on input signal respectively to K frequency band (wherein, K is the natural number more than or equal to 2); The 1st gain arithmetic element, its infer based on said noise spectrum the unit noise spectrum infer the result, calculate said K frequency band noise suppression gain separately; And former sound extra cell; It is in signal after the 1st squelch that after said input signal is used said noise suppression gain, obtains; Add the above input signal with the 1st ratio; Wherein, the 1st ratio is based on that this noise suppression gain and predetermined target noise suppress the difference of gain and definite.
According to the present invention; At first; Because former sound extra cell adds input signal in the signal after the 1st squelch, so, also can carry out what is called to this part and compensate through this former sound even the situation of the excessive reduction of the for example above-mentioned amplitude frequency spectrum that is caused by the noise spectrum presumed value takes place; Therefore, can suppress above-mentioned music generating noise very effectively.
And, in the present invention, owing to have " the 1st ratio ", be based on difference that noise suppression gain and target noise suppress to gain and definite, so the noise suppression effect that in fact obtains is very effective as the meaning of the additional ratio of former sound.Its reason is; Under situation of the present invention; Shown in above-mentioned example, should " the 1st ratio " not that merely the comparison through SNR presumed value and reference value obtains, but according to suppress the relation of gain definite with target noise; In addition, this target noise suppresses to gain and can set based on original expectation noisiness that suppress or that should suppress.
In addition, alleged among the present invention " target noise suppresses gain ", can be through via providing artificially at the operating portion of the outer setting of Noise Suppression Device involved in the present invention etc., perhaps also can be through the computing automatically of any appropriate method.
In Noise Suppression Device involved in the present invention, also can constitute, also have the 2nd gain arithmetic element; It calculates the mean value gain relevant with a said K frequency band, said former sound extra cell to said noise suppression gain; Using to said input signal after the 2nd squelch that obtains after the gain of said mean value in the signal; Add the above input signal with the 2nd ratio, wherein, the 2nd ratio is based on difference that the gain of this mean value and said target noise suppress to gain and definite.
According to this mode, after the 2nd squelch, add input signal with " the 2nd ratio " in the signal.
Here, in this mode, using " mean value gain " is a characteristic.For the gain of this mean value, be meant if suppose above-mentioned noise suppression gain be G (1), G (2) ..., G (K), then for example by Gave=(G (1)+G (2)+... + G (K))/its mean value gain G of calculating ave such as K.In addition; The amplitude frequency spectrum that obtains if further input signal is converted into frequency domain be made as Y (1), Y (2) ..., Y (K), then the output signal of frequency domain can by GaveY (1), GaveY (2) ..., GaveY (K) etc. and obtain (this be alleged in this mode object lesson of " use the 2nd squelch that obtains after the mean value gain after signal ".)。
Therefore, at first, as long as there is the gain of above-mentioned mean value, the situation etc. of the excessive reduction of the for example above-mentioned amplitude frequency spectrum that is caused by the noise spectrum presumed value just can not take place, so can suppress the music generating noise very effectively.
And because in this mode, also the 2nd definite ratio adds former sound to suppress the difference that gains based on above-mentioned mean value gain with above-mentioned target noise, so can more effectively obtain music Noise Suppression effect.
In addition, alleged " after the 2nd squelch, adding input signal with the 2nd ratio in the signal " in this mode, and the relation between above-mentioned " after the 1st squelch, adding input signal with the 1st ratio in the signal " are described below.
At first; Confirm the difference on the concept, that is, the two is relevant with " noise suppression gain " for the generation of " signal after the 1st squelch " and " the 1st ratio " upright surely; The former becomes the object of the former sound additional treatments of carrying out based on the latter; On the other hand, the two is relevant with " mean value gain " for the generation of " signal after the 2nd squelch " and " the 2nd ratio " upright surely, and the former becomes the object of the former sound additional treatments of carrying out based on the latter.
Under this prerequisite; At first; Following in the present invention relation is set up, that is, if alleged " the mean value gain " of this mode of calculating; " former sound extra cell " then involved in the present invention is only additional through carry out former sound based on " the 2nd ratio " of this " mean value gain ", adds and do not carry out former sound through " the 1st ratio ".That is, in the case, the mode of this mode above-mentioned with respect to comprising " the 1st ratio " is in so-called precedence.
Perhaps; Mode such as following is in the present invention also set up; That is, certain part in the input signal is carried out former sound based on " the 1st ratio " add, add (in the case and other parts are carried out former sound based on " the 2nd ratio "; The former part is relevant with " signal after the 1st squelch ", and the latter's part is relevant with " signal after the 2nd squelch ".)。
In addition; The present invention includes mode such as following in its scope in especially; That is, difference is according to circumstances not only carried out former sound to " signal after the 2nd squelch " based on " the 2nd ratio " and is added; And to " signal after the 1st squelch " based on " the 2nd ratio " carry out former sound additional (with aforesaid way as 1 in the variety of way of the present invention, narration in the back.)。Above-mentioned situation promptly comprises following implication; Promptly; Can be to being not to be based on based on " mean value gain " in " signal after the 1st squelch " that " noise suppression gain " set up yet, use based on " mean value gain " and definite " the 2nd ratio " (difference according to circumstances, vice versa.)。
In addition; In Noise Suppression Device involved in the present invention; Also can constitute; Said former sound extra cell calculates the smoothing ratio of the said the 1st or the 2nd ratio after smoothing on the time shaft, after the said the 1st or the 2nd squelch, adds the above input signal with said smoothing ratio in the signal.
According to this mode, calculate the smoothing ratio of the said the 1st or the 2nd ratio after smoothing on the time shaft.Therefore, this smoothing ratio is based on above-mentioned " poor " and determines " the 1st ratio " perhaps " the 2nd ratio ", makes them accept to obtain after smoothing is handled again.In addition; So-called " in smoothing on the time shaft " be meant if suppose after the calculating the smoothing ratio along time sequencing be OGsmt-T (1), OGsmt-T (2) ..., OGsmt-T (r) ... (r is suitable integer); Then for example use suitable smoothing coefficient δ; As OGsmt-T (r)=δ OGsmt-T (r-1)+(1-δ) og etc. and calculate that (wherein, og is " the 1st ratio " perhaps " the 2nd ratio ".)。
According to foregoing, owing to can not make the 1st or the 2nd ratio (the 1st or the 2nd ratio after the smoothing of saying so exactly.That is, " the smoothing ratio " that this mode is alleged) produce fierce variation along with the process of time, so can keep the continuity consistency that squelch is handled.
In addition, in the present invention, as after state shown in; Each frame that preferred pin is divided passing is in time handled, but in the case, alleged " time shaft " in this mode; More particularly, can be assumed to be the axle that under the situation that this frame is arranged singly in order, takes out.About the more detailed object lesson of this point, after describe in embodiment, the particularly formula (7) stated.
In addition, relevant with this mode, alleged " the 1st ratio " perhaps " the 2nd ratio " in this mode, after in the embodiment stated as " former sound adds rate og ", " former sound adds ratio OG in " smoothing ratio " conduct
t", respectively by explanation particularly.
In addition; In the mode of the present invention that comprises above-mentioned notions such as " mean value gains "; Also can constitute, also have the audio detection unit, it through to whether comprising audio frequency in the said input signal detect in time with passing; Thereby this input signal is divided into audio frame that comprises this audio frequency and the noise frame that does not comprise this audio frequency, and signal is in said input signal, to be equivalent to obtain after the said mean value gain of the certain applications of said noise frame after said the 2nd squelch.
According to this mode, the gain of above-mentioned mean value is used to noise frame, more preferably only use to noise frame, thus after generating the 2nd squelch signal (then, signal is accepted the former sound additional treatments based on above-mentioned the 2nd ratio after the 2nd squelch.)。For noise frame, owing to relatively be easy to generate the music noise, this mode then is directed against this problem and the application of the value of averaging gain etc., is used to one of optimal way that obtains the music noise suppression effect so can be described as.
In addition, in this mode, perhaps " do not contain " the such term of audio frequency, can not be interpreted as the so-called absolute meaning for " containing ".For example; The frame that abstractness ground hypothesis " all by audio frequency " occupies and these two kinds of forms of frame of " not having audio frequency "; Two extreme and carry out the situation that " audio frame " reaches the differentiation of " noise frame " to the two; And this mode only also is not limited to and is " noise frame " with the latter, in addition all is the situation that " audio frame " distinguished.Promptly; Even this mode is under the situation of regarding as " noise frame "; Also do not require not comprise " audio frequency " in this noise frame, alleged " audio frame " reaches the differentiation of " noise frame " in this mode, can be that benchmark carries out with the suitable intermediate point of above-mentioned 2 kinds of situation.
In the above-described meaning, perhaps " do not comprise " differentiation that such term or related " audio frame " of this mode reach " noise frame " alleged " comprising " in this mode, can be described as relative.
In this mode, also can constitute, signal is in said input signal, to be equivalent to obtain after the said noise suppression gain of certain applications of said audio frame after said the 1st squelch.
According to this mode, above-mentioned noise suppression gain is used to audio frame, (then, signal is accepted the former sound additional treatments based on above-mentioned the 1st ratio after the 1st squelch to generate signal after the 1st squelch.)。This mode can and be deposited with described mode before, but in the case, preferably only uses the so-called common noise suppression gain of not handling through equalization for audio frame, and only uses the mean value of handling through equalization to gain for noise frame.If in view of the existence of noise in audio frame opposite this situation so obviously and in noise frame, then the related contents processing of this mode can obtain very reasonably efficiently that effective noise suppresses effect.
In this mode, also can constitute, said former sound extra cell replaces said the 1st ratio and adds the above input signal with said the 2nd ratio after said the 1st squelch in the signal.
According to this mode, not only carry out former sound based on " the 2nd ratio " and add to " signal after the 2nd squelch ", also carry out former sound based on " the 2nd ratio " and add to " signal after the 1st squelch ".That is, in the case, to being not to be based on " signal after the 1st squelch " that " noise suppression gain " set up based on " mean value gain ", with based on " mean value gain " and definite " the 2nd ratio " carried out former sound additional treatments.
According to foregoing, the ratio of using as the additional ratio of former sound is integrated with " the 2nd ratio ", thus the high efficiency that realization is handled etc.
In addition; Obtain in the mode of " signal after the 1st squelch " in " audio frame ", using " noise suppression gain "; Also can constitute; Said former sound extra cell will add under the situation of the above input signal in the signal after related said the 1st squelch of said audio frame; Will be to said the 2nd ratio that has calculated with the immediate noise frame of this audio frame, said the 1st ratio as this audio frame adds the above input signal in the signal after said the 1st squelch.
According to this mode, in the former sound additional treatments of audio frame, the former sound that will in the noise frame of its immediate processing, use adds ratio (i.e. " the 2nd ratio "), uses as the additional ratio (i.e. " the 1st ratio ") of its former sound.That is, in the case, its essence is the value of " the 2nd ratio " of immediate noise frame, suppose that but it is the value of " the 1st ratio " of this audio frame.
According to foregoing, in this mode, the squelch in immediate noise frame, carried out is handled, in the squelch of thereafter audio frame is handled, continue to use, from noise frame under the situation that audio frame switches, the consistency that can keep squelch to handle.Thus, under the situation of this switching, can prevent the sharply generation of phenomenon such as variation of noisiness in advance.
In addition, this mode with following content as prerequisite, promptly; Make after the 1st squelch and add former sound with the 1st ratio in the signal, after the 2nd squelch, add the mode of former sound with the 2nd ratio in the signal and deposit, and; The former is relevant with noise frame in the input signal, and the latter is relevant with audio frame.
Perhaps; Obtain in the mode of " signal after the 1st squelch " in " audio frame ", using " noise suppression gain "; Also can constitute; Said former sound extra cell after calculating interim the 2nd ratio, uses the said the 1st or the 2nd ratio of the previous frame of this noise frame will after related said the 2nd squelch of said noise frame, adding under the situation of the above input signal in the signal; Calculating will the smoothing ratio of this interim the 2nd ratio after smoothing on the time shaft; This smoothing ratio as said the 2nd ratio, is added the above input signal in the signal after said the 2nd squelch, after related said the 1st squelch of said audio frame, add under the situation of the above input signal in the signal; With the said the 1st or the 2nd ratio of the previous frame of this audio frame said the 1st ratio, after said the 1st squelch, add the above input signal in the signal as this audio frame.
According to this mode, can suitably confirm to audio frame and noise frame the 1st or the 2nd ratio separately.To noise frame, owing to calculate the smoothing ratio, thereby the consistency continuity that keeps squelch to handle; And, to audio frame, suitably keep in " immediate noise frame " as noted above noise spectrum (promptly; According to this mode, if set the 2nd ratio, then after this to certain noise frame; As long as be continuously audio frame, then with the value of the 2nd ratio, as the value of the 1st new ratio and continue to keep.)。About the more detailed object lesson of this point, after state embodiment, particularly formula (7) time describe.
In addition, in this mode, in the situation that only is called " frame ", comprise the situation of " audio frame ", and the situation of " noise frame ".
In addition, this mode with following content as prerequisite, promptly; Make after the 1st squelch and add former sound with the 1st ratio in the signal, after the 2nd squelch, add the mode of former sound with the 2nd ratio in the signal and deposit, and; The former is relevant with noise frame in the input signal, and the latter is relevant with audio frame.
In addition, relevant with this mode, alleged " the 1st ratio ", " the 2nd ratio " and " smoothing ratio " etc. in this mode, after perhaps " former sound adds ratio OG as " former sound adds rate og " in the embodiment stated
t" wait respectively by explanation particularly, be noted that a kind of appellation and a kind of appellation among the latter among the former in the case, there is not man-to-man corresponding relation.
In addition, in Noise Suppression Device involved in the present invention, also can constitute, the said the 1st or the 2nd ratio is to obtain through following formula (A).
og=max(0,tg-G)…(A)
Here, og is the 1st or the 2nd ratio that should calculate, and tg is that said target noise suppresses gain, and G is said noise suppression gain or the gain of said mean value, max (a, b) expression function that bigger among a and b value is returned.
According to this mode, can suitably calculate the 1st or the 2nd ratio.About the more detailed object lesson of this point, after state embodiment, particularly formula (5) time describe.
On the other hand; In order to solve above-mentioned problem, noise suppressing method involved in the present invention comprises: noise spectrum is inferred operation, and it is based on input signal; Infer the noise spectrum that is included in this input signal respectively to K frequency band (wherein, K is the natural number more than or equal to 2); The 1st gain operational process, its infer based on said noise spectrum operation noise spectrum infer the result, calculate said K frequency band noise suppression gain separately; And former sound additional process; It is in signal after the 1st squelch that after said input signal is used said noise suppression gain, obtains; Add the above input signal with the 1st ratio; Wherein, the 1st ratio is based on that this noise suppression gain and predetermined target noise suppress the difference of gain and definite.
According to the present invention, clear and definite its can realize not having the different action effect of essence with the above-mentioned action effect of recording and narrating to Noise Suppression Device involved in the present invention.
In noise suppressing method involved in the present invention; Also can constitute, also comprise the 2nd gain operational process, it is to said noise suppression gain; Calculate the mean value gain relevant with a said K frequency band; In said former sound additional process,, add the above input signal with the 2nd ratio using to said input signal after the 2nd squelch that obtains after the gain of said mean value in the signal; Wherein, the 2nd ratio is based on difference that the gain of this mean value and said target noise suppress to gain and definite.
According to this mode, clear and definite its can realize in the variety of way with above-mentioned Noise Suppression Device involved in the present invention, to adding that with the 2nd ratio the action effect that the mode of input signal is recorded and narrated does not have the different action effect of essence to the 2nd squelch signal.
In addition, about " after the 2nd squelch, adding input signal with the 2nd ratio in the signal " alleged in this mode relation, same as described above with above-mentioned " after the 1st squelch, adding input signal with the 1st ratio in the signal ".
In addition; In noise suppressing method involved in the present invention; Also can constitute, also comprise the audio detection operation, it through to whether comprising audio frequency in the said input signal detect in time with passing; Thereby this input signal is divided into audio frame that comprises this audio frequency and the noise frame that does not comprise this audio frequency, and signal is in said input signal, to be equivalent to obtain after the said mean value gain of the certain applications of said noise frame after said the 2nd squelch.
According to this mode, clear and definite its can realize that the action effect of recording and narrating in the variety of way with above-mentioned Noise Suppression Device involved in the present invention, to mode from the mean value gain to noise frame that use does not have the different action effect of essence.
In addition, alleged " containing " perhaps " do not contained " meaning of such term in this mode, and be same as described above.
Except that above-mentioned, more specific mode of the present invention or the action effect of realizing being are thus clearly are being recorded and narrated from the record of the embodiment of following beginning.
Description of drawings
Fig. 1 is the block diagram of the structure of the related Noise Suppression Device of expression embodiment of the present invention 1.
Fig. 2 is the process flow diagram of the flow process handled of the related squelch of expression embodiment 1.
Fig. 3 is the key diagram that is used to explain the squelch processed content that embodiment 1 is related.
Fig. 4 is the key diagram that is used to explain existing squelch processed content.
Fig. 5 is the block diagram of the structure of the related Noise Suppression Device of expression embodiment of the present invention 2.
Fig. 6 is the curve map that expression comprises the computing example of the noise suppression gain G (n) under the situation of sound signal.
Fig. 7 is a curve map of representing not comprise the computing example of the noise suppression gain G (n) under the situation of sound signal.
Embodiment
< embodiment 1 >
Below, with reference to Fig. 1, embodiment involved in the present invention 1 is described.In addition, Fig. 1 that here mentions, and in each accompanying drawing of following institute reference (also comprising the for example curve map of Fig. 6 etc.) exists the dimension scale of each several part to compare the situation of appropriate change with physical size.Noise Suppression Device 1 is as shown in Figure 1, infers noise suppression gain operational part 40 between portion 20, noise suppression gain operational part 30, noise period, the additional rate operational part 50 of former sound, former sound additional gain operational part 60, frequency time change portion 70 and audio detection portion 80 by temporal frequency transformation component 10, noise spectrum and constitutes.
The input signal of 10 pairs of time domains of temporal frequency transformation component carries out Fourier transform, and is transformed to the signal of frequency domain.Preferred this Fourier transform is through input signal is passed the frame that is divided into specified quantity in time, and applies suitable window function to this frame and carry out.
The signal of above-mentioned frequency domain is divided into amplitude frequency spectrum and phase frequency spectrum, and the frequency time change portion 70 that phase frequency spectrum is wherein directly stated backward sends.On the other hand, the noise spectrum of amplitude frequency spectrum being stated is backward inferred portion's 20 later each several parts and is carried the various processing of stating after the acceptance.
Also the input signal of above-mentioned time domain is supplied with to audio detection portion 80.Have or not sound signal to detect in 80 pairs of these input signals of audio detection portion.As noted above, input signal is being divided under the situation of framing, to each frame carry out audio detection (in addition, in embodiment 1, with this processing as prerequisite.)。Here, so-called " audio frequency " refers in particular to significant sound for the people such as session, spoken language, music, various signals.That is, below relation is set up, and under the situation that input signal is play through suitable broadcast unit, if play " sound signal " in this input signal, then becomes this sound.
For this sound signal, for example following condition is detected as benchmark, that is, whether the level of input signal surpasses predetermined threshold value.But in addition the present invention can also adopt the whole bag of tricks.For example; The method that can adopt probability of use to add up is inferred the generation probability method of sound signal etc., perhaps, also can adopt and not use above-mentioned input signal; And use signal after its Fourier transform (that is the signal of above-mentioned alleged frequency domain) as method of detected object etc.
In addition, below, will be judged to be the frame that has sound signal through this audio detection portion 80 respectively sometimes and be called " audio frame ", be called " noise frame " with being judged to be the frame that does not have sound signal.In addition, here, exist not have so-called absolute meaning.As noted above, since be that threshold value with regulation is a benchmark, having or not of sound signal judged, so strictness is not precluded within the possibility that comprises the part that is called as sound signal in " noise frame ".
Noise spectrum is inferred portion 20 based on above-mentioned amplitude frequency spectrum, the presumed value of calculating noise frequency spectrum.In embodiment 1, especially,, infer noise spectrum respectively to the frequency band that is split into the regulation number based on following formula (1).
[formula 1]
Here, N
t(n) be the noise spectrum presumed value of certain frame in pre-treatment, N
T-1(n) be the noise spectrum presumed value (therefore, " t " is the subscript of the frame self of expression in pre-treatment) of its former frame, Y (n) is the amplitude frequency spectrum of input, and n is that frequency band is (to the numbering of frequency band mark.In addition, frequency band is split into N.In addition, this N is less than or equal to K in " K frequency band " alleged among the present invention (=N≤K).), β is the smoothing coefficient.In addition, in formula (1), caseA representes that noise spectrum infers the situation that portion 20 handles noise frames, and caseB representes the situation of its process audio frames.
As noted above, noise spectrum infer portion 20 and frame when pre-treatment be noise frame or audio frame accordingly, change is for calculating noise frequency spectrum presumed value N
t(n) formula that uses.That is, when audio frame is handled (caseB), directly use its previous noise spectrum presumed value, obtain noise spectrum presumed value N
t(t), when noise frame is handled (caseA), through the amplitude frequency spectrum smoothing on time shaft that makes input, calculating noise frequency spectrum presumed value N
t(n).
The noise spectrum presumed value N that noise suppression gain operational part 30 is obtained based on above-mentioned amplitude frequency spectrum and through type (1)
t(n), calculating noise suppresses gain.In embodiment 1, especially, suppress gain through following formula (2) calculating noise.
[formula 2]
Here, max (a, b) expression function that the value of bigger that among a and the b is returned (below identical.)。
Through this formula (2), at the amplitude frequency spectrum Y (n) and the pairing noise spectrum presumed value N of input
t(n) between, Y (n)<N
tUnder the situation about (n) setting up, G (n)=0 is at Y (n)>N
tUnder the situation about (n) setting up, G (n)=(Y (n)-N
t(n))/Y (n).
The noise suppression gain that will calculate through this noise suppression gain operational part 30; Respectively with the audio frame of distinguishing through above-mentioned audio detection portion 80 and noise frame accordingly, via noise suppression gain operational part between noise period 40 or directly supply with to former sound additional gain operational part 60.Noise Suppression Device 1 shown in Figure 1 has the switch (with reference to the camber line arrow of figure) that is used for realizing above-mentioned processing.
Noise suppression gain operational part 40 between noise period (below, for simply, be also referred to as " using gain operational part 40 between noise period ".), calculating should be to the noise suppression gain of noise frame application.In embodiment 1, adopt following method in order to calculate this noise suppression gain.
At first, the noise suppression gain G (n) based on through type (2) is obtained calculates the g by following formula (3) expression.
[formula 3]
Clearly can know the mean value relevant of the noise suppression gain of this g expression (2) with frequency band n from the right of formula (3).
Next, through following formula (4), make the noise suppression gain mean value g smoothing of this formula (3).
[formula 4]
G
t=μG
t-1+(1-μ)g……(4)
Here, μ is the smoothing coefficient, G
tBe noise suppression gain to the noise frame in pre-treatment, G
T-1It is noise suppression gain to the noise frame of its previous processing.
The formula that illustrates as caseA in the above-mentioned formula (1) also is like this; But in this formula (4); Because when the noise suppression gain of calculating when the frame in the pre-treatment; With reference to the noise suppression gain of the frame of its previous processing, thus we can say carried out on the time shaft smoothing (for after the formula (7) stated also identical.)。
G in this formula (4)
t, be should be between this noise period with obtain in the gain operational part 40, be used for to the noise suppression gain of using between noise period (below, for simply, be also referred to as " between noise period with gain ".)。
Use gain G between the noise period that will obtain according to aforesaid way with gain operational part 40 between noise period
t, unification is applied to whole frequency bands.Below, in order to represent this situation, with this unified G that uses
t, be expressed as G1 (n).In the case, G1 (0), G1 (1) ..., G1 (N-1), equal G all
t
The additional rate of former sound of the former tone signal that the signal after additional rate operational part 50 calculating of former sound and the squelch is relative.In embodiment 1, especially, calculate the additional rate og of this former sound based on following formula (5).
[formula 5]
og=max(0,tg-G
t)……(5)
Here, tg is that target noise suppresses gain, is based on following formula (6) and obtains.
[formula 6]
TG in this formula (6) is the target noise amount of suppression, is unit with dB.This TG (perhaps tg) is through outside provide via not shown operating portion etc. from installing artificially, perhaps, and also can be through the computing automatically of any appropriate method.
According to above formula (5), suppress to use gain G between gain tg and noise period at target noise
tBetween, tg<G
tUnder the situation about setting up, og=0 is at tg>=G
tUnder the situation about setting up, og=tg-G
t
Former sound additional gain operational part 60 calculates the noise suppression gain after former sound adds based on the additional rate og of above-mentioned former sound.In embodiment 1, adopt following method in order to calculate this noise suppression gain.
At first, the additional rate og of former sound based on through type (5) is obtained calculates the OG by following formula (7) expression
t
[formula 7]
Here, OG
tIt is the additional ratio of former sound of the frame in pre-treatment.OG
T-1Be the additional ratio of former sound of its previous frame, λ is the smoothing coefficient.In addition, caseA in the formula (7) and the implication of caseB are identical (also identical in following formula (8) with the situation of above-mentioned formula (1).)。
As noted above, former sound additional gain operational part 60 and frame when pre-treatment be noise frame or audio frame accordingly, change is in order to calculate the additional ratio OG of former sound
tAnd the formula that uses.That is, when audio frame is handled (caseB), directly use the additional ratio of its previous former sound to calculate the additional ratio OG of former sound
t, when noise frame is handled (caseA), calculate the additional ratio OG of former sound through making the additional rate og smoothing on time shaft of above-mentioned former sound
t
Next, former sound additional gain operational part 60 calculates the noise suppression gain after former sound adds based on following formula (8).
[formula 8]
Here, shown in G1 (n) is as described above, be illustrated in the noise frame using gain between the unified noise period of using of all frequency bands.
According to this formula (8), with situation in the above-mentioned formula (7) respectively accordingly, calculate the noise suppression gain G2 (n) of former sound after additional (below, for simply, be also referred to as " revising back gain G 2 (n) ".)。
Multiplier 11 shown in Figure 1 makes that gain G 2 (n) multiplies each other with amplitude frequency spectrum Y (n) after the correction of obtaining according to aforesaid way.That is, carry out the computing of S (n)=G2 (n) Y (n), consequently, the amplitude frequency spectrum S (n) after the squelch that can obtain finally should obtaining.
At last, frequency time change portion 70 based on the amplitude frequency spectrum S (n) after the squelch of obtaining according to aforesaid way with from the temporal frequency transformation component 10 direct phase frequency spectrums of supplying with, generates the output signal of time domain.In embodiment 1, owing in temporal frequency transformation component 10, carry out Fourier transform, frequency time change portion 70 implements inverse Fourier transform.
Below, on the basis with reference to the Fig. 1 that crosses,, effect, action and the effect of the Noise Suppression Device 1 that above-described embodiment 1 is related is described with reference to Fig. 2 to Fig. 4.
At first, 10 pairs of input signals of temporal frequency transformation component are implemented Fourier transform, and with its amplitude frequency spectrum Y (n) and phase frequency spectrum (the step S101 of Fig. 2) of being decomposed into as shown in Figure 1.At this moment, temporal frequency transformation component 10 is as noted above, implements to handle to each frame.
In addition, parallel therewith, whether comprise sound signal in 80 pairs of input signals of audio detection portion and detect (the step S102 of Fig. 2).It is the processing that can input signal be divided into audio frame and noise frame that this detection is handled.The processing that audio detection portion 80 also is correlated with.
Then, noise spectrum is inferred portion 20 through above-mentioned amplitude frequency spectrum Y (n) and above-mentioned formula (1), to each frequency band n with Rack, calculates noise spectrum presumed value N
t(n).In the case, as noted above, with when the frame of pre-treatment be noise frame or audio frame accordingly, carry out different processing (with reference to the step S103 of Fig. 2).In addition, as shown in Figure 2, from this noise spectrum presumed value N
t(n) after the computing, during till the output signal generation processing of being undertaken by multiplier shown in Figure 1 11 (the step S104 of Fig. 2), and the difference between noise frame and the audio frame accordingly, launches the different processing of its content essence.Therefore, below, first, the processing used to noise frame, second, to the processing that audio frame is used, be divided into [I] and [II] and describe.
In addition, above-mentioned processing respectively is as shown in Figure 1, is through carrying out with the switching of the testing result corresponding switch of audio detection portion 80.
[I] at first, at noise frame with in handling, through in the above-mentioned formula (1) as the formula shown in the caseA, calculate noise spectrum presumed value N
t(n) (the step S201 of Fig. 2).As noted above, this is to handle through the smoothing of the amplitude frequency spectrum Y (n) of input to carry out.
Then, based on above-mentioned noise spectrum presumed value N
t(n) and above-mentioned formula (2), calculate noise suppression gain G (n) (the step S202 of Fig. 2).This is to carry out through the effect of the noise suppression gain operational part 30 among Fig. 1.As noted above, at Y (n)>N
tUnder the situation about (n) setting up, then become G (n)=(Y (n)-N
t(n))/and Y (n), under its invalid situation, then become G (n)=0.Thus, for example can obtain the noise suppression gain shown in Fig. 3 (C) (in addition, respectively in Fig. 3 (B) illustration go out above-mentioned noise spectrum presumed value N
t(n), illustration goes out the amplitude frequency spectrum of input signal in Fig. 3 (A).)。
Then,, obtain the mean value g relevant of this noise suppression gain G (n), and handle, calculate and use gain G between noise period through the smoothing that is directed against this g with frequency band through above-mentioned formula (3) and formula (4)
t(the step S203 of Fig. 2).This process equalization smoothing and use gain G between the noise period that obtains
t, become the G1 (n) of all band sharings.This is through carrying out with the effect of gain operational part 40 between noise period.
As noted above; In embodiment 1; The big characteristic of one of which is; Not that the noise suppression gain G (n) that through type (2) is obtained directly uses under previous status, but will carry out using gain G between the noise period after the equalization on the time shaft of the equalization relevant and formula (4) of formula (3) to this G (n) with frequency band
t, use with gain G 1 (n) between the noise period of using as all frequency bands.
In addition, in Fig. 3 (D), illustration goes out to an example of the situation of averaging of noise suppression gain G (n) processing (also can with reference to the dotted line shown in Fig. 3 (C)).
Then, through using gain G between above-mentioned noise period
tAnd above-mentioned formula (5), calculate the additional rate og (the step S204 of Fig. 2) of former sound.This is to carry out through the effect of the additional rate operational part 50 of the former sound among Fig. 1.Here, the setting of target noise inhibition gain tg and even target noise amount of suppression TG how, works as a main governing factor.That is, if use gain G between noise period
tSuppress gain tg with target noise and compare greatly, then the additional rate og of former sound is set at 0,, then use gain G between setting and noise period if not like this
tAdditional rate og (that is og=tg-G, of corresponding former sound
t).The meaning of both of these case separate processes is, confirms the relation of basis and target noise amount of suppression TG, obtain through adding the effect of the tone quality improving that former sound brings with which kind of degree.That is, in the latter case, with (that is, tg and G in the frame of confirming through the target noise amount of suppression
tThe pairing part of difference) add former sound, the realization tone quality improving is a fundamental purpose, under the former situation, because G
t>tg sets up, and the leeway of not improved tonequality is so be made as 0 (in the case, having suppressed noisiness thus on the contrary increases) with the additional rate og of former sound.As a result, there are the meaning that realizes following processing in above-mentioned formula (5) and formula (6), that is, be benchmark to observe the target noise amount of suppression, in addition, exist under the situation in the leeway that adds former sound, in this frame, realize the improvement of tonequality.
As noted above, in embodiment 1, the big characteristic of one of which is that the additional rate og of former sound uses gain G through using between noise period
tObtain.
Then, through in additional rate og of above-mentioned former sound and the above-mentioned formula (7) as the formula shown in the caseA, calculate the additional ratio OG of former sound
t(the step S205 of Fig. 2).This former sound adds ratio OG
tAs noted above, through the additional rate og smoothing on time shaft of former sound is obtained.Then, through the additional ratio OG of the former sound of obtaining according to aforesaid way
tAnd above-mentioned formula (8), calculate the noise suppression gain after former sound adds, promptly revise back gain G 2 (n).More than be to carry out through the effect of former sound additional gain operational part 60.
In the case, this correction back gain G 2 (n) finally has following meaning, that is, be the gain of on the basis of the degree of adding with gain G 1 (n) and former sound between the noise period of having considered the smoothing of above-mentioned process equalization, confirming.
In addition, situation about preferably will install after the starting is just taken into account, suitably confirms the N in conduct and the above-mentioned formula (1)
T-1(n) initial value of corresponding value is (as the N of above-mentioned initial value
T-1(n), certainly, also can after the audio frame stated with the noise spectrum presumed value N in handling
t(n) use in the computing.)。G in above-mentioned formula (4), the formula (7)
T-1(n) also identical therewith.
[II], carries out basically with roughly the same each of above-mentioned noise frame processing and handles with in handling at audio frame on the other hand.That is, calculate noise spectrum presumed value N
t(n) and based on its noise suppression gain G (n) (with reference to the step S301S202 of Fig. 2), based on the additional ratio OG of former sound
tCalculate and revise back gain G 2 (n) (the step S303S304 of Fig. 2), these handle identical with noise frame.
But, in this audio frame is handled, compare with the noise frame processing, there are the different or lime lights of the following stated.
(i) noise spectrum presumed value N
t(n) not to obtain through the formula in the above-mentioned formula (1) as shown in the caseA, but (the step S301 of Fig. 2) through obtaining as the formula shown in the caseB.Because this formula is N
t(n)=N
T-1(n), so we can say that it is the processing that what is called is maintained the statusquo that audio frame is handled.A little at length saying, is under the situation of noise frame before this audio frame again, the noise spectrum presumed value N that will calculate this noise frame
T-1(n), directly in this audio frame is handled, use, in addition, on the other hand, be audio frame before this audio frame and before the audio frame before this, be under the situation of noise frame, the noise spectrum presumed value N that will calculate this noise frame
T-2(n), directly, uses this audio frame in handling etc.
In a word, in audio frame, use the noise spectrum presumed value N that immediate noise frame is calculated
T-p(n) (p is that the previous frame from this audio frame begins number, the frame number (comprising two ends) till this immediate noise frame).
(ii) state the additional ratio OG of former sound that formula (7) is carried out in the use
tComputing in, also can be described as identical therewith.That is, since formula (7) in be OG as the formula shown in the caseB
t=OG
T-1So, in the case, in audio frame is handled, or what is called maintain the statusquo (with reference to the step S303 of Fig. 2).
If make expression consistent, then in audio frame, use the additional ratio OG of the former sound that immediate noise frame is calculated with above-mentioned situation
T-p(n) (p begins number, the frame number (comprising two ends) till this immediate noise frame for the previous frame from this audio frame).
The (iii) computing of noise suppression gain G (n) itself, irrelevant with differentiation audio frame and noise frame, use above-mentioned formula (2) to carry out identically.In the step S202 of Fig. 2, will be connected with [audio frame is with handling] corresponding frame with [noise frame is with handling] and describe, be in order to represent this characteristic (still, N in the formula (2) with signifying
t(n) value, respectively with formula (1) in caseA and caseB corresponding, that yes is different for two kinds of frames.)。
(iv) in audio frame is handled, do not carry out the processing relevant, promptly to the equalization smoothing processing of noise suppression gain G (n) (with reference among the step S203 of Fig. 2 and this figure right-hand) with above-mentioned formula (3) and formula (4).In addition, accompany therewith, use gain G owing to become so-called the existence between proper noise period
tState, so do not carry out the processing relevant, promptly calculate the processing (with reference among the step S204 of Fig. 2 and this figure right-hand) of the additional rate og of former sound with above-mentioned formula (5) yet.
(gain G 2 (n) after the correction that v) finally calculates is not to obtain through the formula as shown in the caseA in the above-mentioned formula (8), but (the step S304 of Fig. 2) through obtaining as the formula shown in the caseB.In the case, difference is, when noise frame is handled, use through between the noise period of equalization smoothing with gain G 1 (n), and when audio frame is handled, the noise suppression gain G (n) that direct use is obtained according to formula (2).
Through the processing of above [I] and [II], can obtain revising back gain G 2 (n), if make this corrections afterwards gain G 2 (n) multiply by original amplitude frequency spectrum Y (n), then calculate the amplitude frequency spectrum S (n) (the step S104 of Fig. 2) after the squelch.
In Fig. 3 (E), for simply, the amplitude frequency spectrum Y (n) that makes Fig. 3 (A) is shown merely, multiply by noise suppression gain (that is result who, g) obtains after the equalization of Fig. 3 (C).In embodiment 1, as noted above, in addition, also consider the additional degree of former sound in interior gain adjustment (with reference to formula (8), particularly OG
t(n) effect), but the essence of the processing that Fig. 3 (E) fully expresses hypothesis when omitting the situation to the consideration of above-mentioned former sound additional treatments (in formula (8), only show as, if OG
t(n)=0, revise back gain G 2 (n) and just equal G1 (n) or G (n) simply.)。
According to the Noise Suppression Device with above-described structure and effect 1, can obtain following effect.
At first, according to the Noise Suppression Device in the embodiment 11, can very suitably suppress to be included in the noise in the input signal.Here, in alleged " suitably ", with regard to embodiment 1, the each item that comprises following record especially is true.
(1) first, according to embodiment 1, can prevent so-called music generating noise very effectively.Here, so-called music noise is meant from the amplitude frequency spectrum of input signal, deducting the noise that produces after the noise spectrum presumed value.
For example; Noise suppression gain based on the noise spectrum presumed value; Can use (Y (n)-N (n))/Y (n) in the above-mentioned formula (2) simply and obtain; If suppose state that it is directly used in multiplier shown in Figure 1 11, then the amplitude frequency spectrum S (n) after the squelch is obtained as S (n)={ (Y (n)-N (n))/Y (n) } Y (n)=Y (n)-N (n).That is, in the case,, obtain the amplitude frequency spectrum S (n) after the squelch through merely from the amplitude frequency spectrum of input signal, deducting the noise spectrum presumed value.
But, because the noise spectrum presumed value under this situation is nothing but " presumed value ", so the noise spectrum of not necessarily reflection reality.Therefore, exist in certain frequency band, deduct the situation of going back residual noise after the noise spectrum presumed value; In addition, also exist in other frequency bands, produce the excessive situation of reduction and (cut down under the excessive situation at this; Owing to there is not negative amplitude frequency spectrum, be 0 so be set.)。In Fig. 4; Above-mentioned situation generalities ground is represented for example, the solid line among Fig. 4 (C) is represented to deduct the back and also had residue (with reference to label " KN "); Dotted line representes to cut down excessively each situation of (with reference to label " HS ") (in addition, Fig. 4 (A) reaches (B) and Fig. 3 A) and (B) identical.In addition, the part that the label HSt of Fig. 4 (C) is referred to is the illustration of Y (n)-situation that N (n)=0 by chance sets up.)。
If above-mentioned amplitude frequency spectrum S (n) is carried out inverse Fourier transform to time domain, then this signal becomes the synthetic signal of sine wave that has random frequency by a plurality of, if it is play, then can hear very ear-piercing sound.Here it is music noise.
As noted above, the main cause of generation music noise, strictness is that the noise spectrum and the noise spectrum presumed value of unknowable reality is inconsistent.
In embodiment 1, suppressed above-mentioned music generating noise very effectively.Its reason is when noise frame is handled, to use gain G between the noise period after the smoothing of use equalization
tCalculate and revise back gain G 2 (n), and be applied to amplitude frequency spectrum Y (n) (with reference to Fig. 3 (E)).Thus, under the state of keeping the original frequency structure that amplitude frequency spectrum had, carry out squelch, so the music noise is given birth in utmost point difficult labour.
(1-i) in addition, during calculating noise, use gain G
tThe equalization of Shi Jinhang (above-mentioned formula (3)) and smoothing (above-mentioned formula (4)) have intrinsic meaning respectively.The former purpose clearly can be known from Fig. 3, mainly is to obtain above-mentioned music Noise Suppression effect, and the latter's purpose mainly is the angle of lasting from what is called, keeps the continuity that squelch is handled.According to the latter, owing to use gain G between noise period
t(n) rapid variation can not take place along with effluxion; So under situation the about for example signal that is included in this noise frame being play; Can not cause discomfort (smoothing processing (that is caseA in the formula (1), of other that in embodiment 1, carry out in addition, acoustically to the listener; CaseA in the formula (7)), has the identical meaning of essence therewith in principle.)。
(2) second, above-mentioned (1) is to handle relevant music generating noise about prevention with noise frame, is associated therewith, according to embodiment 1, can realize more reliably preventing handling relevant music generating noise with audio frame.This is as noted above; Through in audio frame is handled; Directly use to calculate with what is called and revise back gain G 2 (n) (caseB in the formula (8), perhaps above-mentioned [II] (v)) and realizing without the form of the noise suppression gain G (n) (with reference to formula (2)) of equalization smoothing.
(3) and, the 3rd, according to embodiment 1, from noise frame under the situation that audio frame switches, the consistency that can keep noise suppression to handle.This is as noted above, through when audio frame is handled, as noise spectrum presumed value N
t(n), use the N that immediate noise frame is calculated
T-p(n) (with reference to above-mentioned [II] record (i)) and realize.
Conclude above (2) and (3), in embodiment 1, time has been descended in following aspect; Promptly; Not only in audio frame, carry out effective noise and suppress, the squelch when also respecting the noise frame processing is in addition handled (particularly its effect), thereby makes two kinds of conversions between the frame become more natural.Thus, with under the Noise Suppression Device in the embodiment 11 and the situation that certain audio playing unit is connected, from noise frame under the situation that audio frame switches, can not cause the discomfort acoustically such as variation of the related volume sense of noise to the listener.
In addition, in order to suppress the music noise in the audio frame, also consider following method, promptly; Replace above-mentioned S (n)=Y (n)-N (n), and adopt S (n)=Y (n)-α N (n), and make the value of this α (>0) bigger; But wherein there is following shortcoming, that is, makes the possibility of tonequality rapid deterioration very high.But,, then make the music Noise Suppression insufficient if reduce α.
In addition; Consider also that through adopting following method to suppress the music noise promptly, the amplitude frequency spectrum after that the dotted line in Fig. 4 (C) is represented, the squelch becomes 0 part (promptly; The part that label HS and HSt are referred to) in, adds fixing value (noise floor (noise floor)).This is to come from idea; That is, want this section H S and HSt to be increased, cover remainder KN after deducting (perhaps make its not obvious) (in addition through what is called; Because under situation with this method and the method for using above-mentioned α and usefulness; Can set α littlelyr, so in the case, also can be prevented the effect of sound quality deterioration.)。
But; Because additional above-mentioned noise floor just means that the absolute magnitude that makes noise increases, so not only from reaching the angle existing problems that suppress this original purpose of noise; And have difference along with the setting of this noise floor amount, make the noise suppression effect extremely inadequate problem that becomes probably.
From above-mentioned angle, the Noise Suppression Device 1 in also can clear and definite embodiment 1 has advantage very much.Promptly; In embodiment 1, owing to not as using above-mentioned α, the amount that deducts is mechanically increased; So possibly produce sound quality deterioration hardly; In addition, owing to be not simple additional such processing of carrying out above-mentioned noise floor, so the situation of the effect of the squelch that does not have sacrifice and carried out.In addition, as recording and narrating, even so, also can suppress the music noise effectively.
(4) according to the Noise Suppression Device in the embodiment 11 because as the step S205 of reference above-mentioned formula (5)~formula (7) or Fig. 2 and S303 carry out former sound additional treatments illustrated, so can more effectively realize noise suppression effect.According to this former sound additional treatments; Owing to can expect the effect identical with the additional treatments of above-mentioned noise floor; Promptly; To the effect of covering of the residual fraction KN after the deducting among Fig. 4 (C), so can more effectively suppress the music noise, or prevent that (still, above-mentioned noise floor is " fixing " all the time for the deterioration of tonequality.This point be with the situation of utilizing " former sound " between conclusive different.)。
In addition; In above-mentioned; In order to hold the effect that realizes through the Noise Suppression Device in the embodiment 11 more clearly; With the method for using above-mentioned α or use during the method for noise floor compares, have the part that this effect is described, but the present invention does not have and gets rid of these energetically and use α or noise floor to suppress the intention of the method for music noise.That is, can these methods and the present invention and various forms and usefulness also be used mode according to above-mentioned, can draw the advantage of this method, simultaneously, make the effect of the present invention and variety of way thereof more remarkable.
And, in embodiment 1, not to carry out former sound simply to add, have following some characteristic.
(4-i) at first, additional ratio (that is OG, of former sound
t) be based on the additional rate og of former sound and definite, the additional rate og of this former sound be and noise period between use gain G
tSuppress with target noise that the size cases of gain tg confirms accordingly.Specifically, as recording and narrating, because in former sound additional treatments; Will be (promptly as the degree of the squelch of target; Tg) as a main governing factor, through and relation between it confirm that former sound adds rate og, so through during based on noise, using gain G
tProcessing and former sound additional treatments between, use evenly, can more effectively obtain noise suppression effect and music noise suppression effect, even the tone quality improving effect.
(4-ii) in addition, for above-mentioned former sound additional treatments, when audio frame is handled, as the additional ratio OG of former sound
t, use the OG that immediate noise frame is calculated
T-p(with reference to above-mentioned [II] record (ii)).The essence of this practice with as the noise spectrum presumed value N of certain audio frame
t(n), directly use the noise spectrum presumed value N of its previous frame
T-1(n) so above-mentioned idea is identical.That is, in this former sound additional treatments, under the situation of the switching between noise frame and the audio frame, the consistency that keeps squelch to handle.
< embodiment 2 >
Below, with reference to Fig. 5 to Fig. 7, embodiment involved in the present invention 2 is described.In addition, this embodiment 2, with the contrast of above-mentioned embodiment 1 in, exist with audio detection and handle relevant difference, and for other aspects, otherwise explain especially, then identical with above-mentioned embodiment 1.Therefore, below, being primarily aimed at above-mentioned difference and describing, simplification even omission are to otherwise explanation.In addition, for the label on the accompanying drawing, except above-mentioned difference, continue to use label.
Noise Suppression Device 1 ' in this embodiment 2 is as shown in Figure 5, has the structure that audio detection portion 801 is connected the back segment of noise suppression gain operational part 30.That is, this audio detection portion 801 is detected having or not sound signal in the input signal through using the noise suppression gain G (n) that is calculated by above-mentioned formula (2), perhaps audio frame and noise frame is distinguished.
In embodiment 2, for to having or not sound signal to detect, and adopt following method.
At first, based on the noise suppression gain G (n) that through type (2) is obtained, calculate Var by following formula (9) expression.
[formula 9]
Here; G is the g through formula (3) expression of in above-mentioned embodiment 1, using, and generally speaking, is that the mean value relevant with frequency band n to G (n) is (in embodiment 2; Not only pass through to use gain operational part 40 between noise period, and carry out the computing of this g through audio detection portion 801.Certainly, also can be with the operation result that draws by the side among both, shared between the two.)。
Can know clearly that from formula the Var of this formula (9) representes the variance of G (n).
Then, whether this Var being surpassed setting judges.The meaning of this judgement is following.
Usually, the noise suppression gain G (n) that through type (2) calculates comprising under the situation of sound signal and not comprising under the situation of sound signal, shows different form.Fig. 6 and Fig. 7 represent an one of which example, and the former is the computing example that comprises the noise suppression gain G (n) under the situation of sound signal, and the latter is the computing example that does not comprise the noise suppression gain G (n) under the situation of sound signal.These figure are compared clearly and can know,, infer easily that then between the two produces bigger this conclusion of gap if calculate the variance of the G separately (n) under two kinds of situation.That is, can carry out to judge with the reliability of certain degree, that is, arrive to a certain degree greatly if be directed against the variance yields of the G (n) of certain frame, then it comprises sound signal, if not like this, does not then comprise sound signal.
The meaning of judging to the size of above-mentioned Var just is this.In other words, suppose to have certain setting VB,, therefore, it is divided into " audio frame ",, therefore, it is divided into " noise frame " if Var≤VB does not then have sound signal in this frame if then there is sound signal in Var>VB in this frame.
In addition, different with the structure of Fig. 1 in the structure of Fig. 5, noise spectrum is inferred the testing result that portion 20 can't use sound signal.That is, it is not prerequisite to distinguish audio frame and noise frame that noise spectrum is inferred portion 20, and calculating noise frequency spectrum presumed value N
t(n).
Noise spectrum presumed value N under the above-mentioned situation
t(n), can obtain through formula (10) and formula (11) for example.
[formula 10]
PA
t(n)=αPA
t-1(n)+(1-α)Y(n)……(10)
[formula 11]
Here, PA
t(n) being amplitude frequency spectrum in the input signal in the frame in the pre-treatment, is the amplitude frequency spectrum after the smoothing, PA
T-1(n) being this amplitude frequency spectrum in its previous frame, is the amplitude frequency spectrum after the smoothing, and α is the smoothing coefficient, and γ β is a controlled variable.In addition, in formula (11), caseC representes PA
t(n)>N
T-1Situation about (n) setting up, caseD representes situation in addition.
In the case, the formula of representing as caseD in the formula (11) and the combination of formula (10), in fact with above-mentioned formula (1) in the formula of representing as caseA synonym roughly.
On the other hand, the formula of representing as caseC in the formula (11), be not in the above-mentioned formula (1) pairing that.But, because this formula is as noted above, that is, and PA
t(n)>N
T-1Under the situation about (n) setting up; That is to say; Under the situation of amplitude frequency spectrum above the noise spectrum presumed value of its previous frame of the frame in pre-treatment, play a role; So be not cannot think this caseC can hint out the frame that deserves in the pre-treatment be audio frame possibility (if for a large amount of n (=0,1,2,3 ...) satisfy above-mentioned condition, we can say that then its possibility is higher.But, be " hint " after all.)。
We can say that these formulas (10) and formula (11) have intercommunity with above-mentioned formula (1) in above-mentioned meaning scope.
In a word, it is constant suitably calculating noise spectrum presumed value this point.
As noted above, can obtain following effect according to embodiment 2.
At first, can clearly pass through this embodiment 2, also can obtain not having the action effect of essential distinction with the action effect that obtains through above-mentioned embodiment 1.That is, in this embodiment 2, can obtain with record and narrate to above-mentioned embodiment 1 from the roughly the same effect of (1) to the effect of (4).
And, Fig. 1 and Fig. 5 are compared clearly and can know, can obtain the raising of treatment effeciency, the effects such as simplification that circuit constitutes according to this embodiment 2.This be because, the audio detection in the embodiment 1 is so-called to carry out independently, replaces it, the audio detection in the embodiment 2 is carried out through utilizing noise suppression gain G (n), has carried out its subordinateization.
In the present invention, because the computing of noise suppression gain G (n) is the processing that must carry out, handle so utilize its operation result to carry out audio detection, the high efficiency that nature can obtain to handle is rationalized.And it detects performance quite high (with reference to comparison diagram 6 and Fig. 7).
More than, embodiment involved in the present invention is illustrated, but Noise Suppression Device involved in the present invention is not limited to aforesaid way, can carry out various distortion.
(1) in above-mentioned embodiment 1 and 2, makes and use gain G between noise period
tEqualization on frequency axis, smoothing on time shaft, but the present invention is not limited to aforesaid way.Narrate in the above, handle and the smoothing processing,,, can according to circumstances omit so particularly handle for smoothing because its fundamental purpose is different for equalization.Shown in Fig. 3 (E), handle even only implement equalization, also can obtain music Noise Suppression effect to a certain extent.
(2) in addition, in above-mentioned embodiment 1 and 2, use gain G between noise period
tHandle and the smoothing of above-mentioned formula (4) is handled and obtained through the equalization of above-mentioned formula (3), but the present invention and be not limited to above-mentioned formula (3) and the form of formula (4).
At first, in the present invention, noise suppression gain mean value g is not limited to the form that through type (3) calculates.
That is, though in formula (3), g be to use whole N frequency band (all be N, that is, the 0th, 1,2 ..., a N-1 frequency band) calculate, but this g for example also can only use a part of frequency band wherein and calculate.In the case, can consider to use except the two perhaps frequency band the side of extremely low frequency band (near the frequency band of DC composition) and EHF band (near being the frequency band of Qwest's frequency).
In addition, when calculating noise suppresses gain mean value g, can carry out different weightings to each frequency band.For example, only make specific frequency band multiply by certain specific weighting coefficient, perhaps, the weighting coefficient that makes all frequency bands multiply by continuously, interimly increase or reduce, or the like.
In addition, in the present invention, use gain G between noise period
tBe not limited to the form calculated through above-mentioned formula (4).
That is, though in formula (4), G
tThrough noise suppression gain mean value g smoothing on time shaft is obtained, but this G
tAlso can be used as for example adjacent frame g mean value and calculate.
(3) and, in above-mentioned embodiment 1 and 2, though use gain G between noise period with the equalization smoothing
tReach G1 (n) and be applied to all frequency bands (with reference to the caseA in the formula (8) or Fig. 3 (E)), but the present invention is not limited to aforesaid way.
For example, can be only to above-mentioned except extremely low frequency band and EHF band the two or the frequency band the side, use between this noise period and use gain G
tAnd G1 (n).In the case, for the frequency band of getting rid of beyond this range of application, gain that can the application of fixed value.
(4) in above-mentioned embodiment 1 and 2, noise suppression gain G (n) is that through type (2) calculates, but the present invention is not limited to aforesaid way.For example, in addition, can use S filter method, MMSE (Minimum Mean-Square Error) method etc. (about these methods, with reference to above-mentioned non-patent literature 3 and 4).Also can infer SNR (audio frequency (signal)/noise proportional), suppress gain G (n) based on this SNR calculating noise.
(5) in above-mentioned embodiment 2, in order between audio frame and noise frame, to distinguish, and obtain the variance on frequency axis of noise suppression gain G (n), but the present invention is not limited to aforesaid way through above-mentioned formula (9).
For example, certainly replace variance and use standard deviation, also can service time variance or standard deviation on the axle.In addition, also can be based on among the noise suppression gain G (n) of each frequency band, have and how much drop in the space that 2 reference values by regulation are divided into etc.; The differentiation of carrying out between audio frame and the noise frame (for example, if its quantity is bigger, then can be judged as noise suppression gain G (n) and concentrates and to be present on certain position; So we can say that the degree of its dispersion is less; Therefore, this frame is identified as noise frame, or the like.)。In addition, above-mentioned various determination methods can according to circumstances and also be used.Thus, for example, with reference on above-mentioned variance and the two the basis of quantity of dropping on the noise suppression gain G (n) in the above-mentioned space, the degree of disperseing judged etc.
Claims (14)
1. Noise Suppression Device is characterized in that having:
Noise spectrum is inferred the unit, and it infers the noise spectrum that is included in this input signal based on input signal respectively to K frequency band, and wherein, K is the natural number more than or equal to 2;
The 1st gain arithmetic element, its infer based on said noise spectrum the unit noise spectrum infer the result, calculate said K frequency band noise suppression gain separately; And
Former sound extra cell; It is in signal after the 1st squelch that after said input signal is used said noise suppression gain, obtains; Add the above input signal with the 1st ratio; Wherein, the 1st ratio is based on that this noise suppression gain and predetermined target noise suppress the difference of gain and definite.
2. Noise Suppression Device as claimed in claim 1 is characterized in that,
Also have the 2nd gain arithmetic element, it calculates the mean value gain relevant with a said K frequency band to said noise suppression gain,
Said former sound extra cell; Using to said input signal after the 2nd squelch that obtains after the gain of said mean value in the signal; Add the above input signal with the 2nd ratio, wherein, the 2nd ratio is based on difference that the gain of this mean value and said target noise suppress to gain and definite.
3. Noise Suppression Device as claimed in claim 1 is characterized in that,
Said former sound extra cell,
Calculate the smoothing ratio of said the 1st ratio after smoothing on the time shaft,
After said the 1st squelch, add the above input signal with said smoothing ratio in the signal.
4. Noise Suppression Device as claimed in claim 2 is characterized in that,
Said former sound extra cell,
Calculate the smoothing ratio of said the 2nd ratio after smoothing on the time shaft,
After said the 2nd squelch, add the above input signal with said smoothing ratio in the signal.
5. Noise Suppression Device as claimed in claim 2 is characterized in that,
Also have the audio detection unit, it is through to whether comprising audio frequency in the said input signal detecting with passing in time, thereby this input signal is divided into audio frame that comprises this audio frequency and the noise frame that does not comprise this audio frequency,
Signal is in said input signal, to be equivalent to obtain after the said mean value gain of the certain applications of said noise frame after said the 2nd squelch.
6. Noise Suppression Device as claimed in claim 5 is characterized in that,
Signal is in said input signal, to be equivalent to obtain after the said noise suppression gain of certain applications of said audio frame after said the 1st squelch.
7. Noise Suppression Device as claimed in claim 6 is characterized in that,
Said former sound extra cell replaces said the 1st ratio and adds the above input signal with said the 2nd ratio after said the 1st squelch in the signal.
8. Noise Suppression Device as claimed in claim 6 is characterized in that,
Said former sound extra cell,
To after related said the 1st squelch of said audio frame, add under the situation of the above input signal in the signal,
Will be to said the 2nd ratio that has calculated with the immediate noise frame of this audio frame, said the 1st ratio as this audio frame adds the above input signal in the signal after said the 1st squelch.
9. like claim 6 or 8 described Noise Suppression Devices, it is characterized in that,
Said former sound extra cell,
To after related said the 2nd squelch of said noise frame, add under the situation of the above input signal in the signal,
After calculating interim the 2nd ratio; Use the said the 1st or the 2nd ratio of the previous frame of this noise frame; Calculating will the smoothing ratio of this interim the 2nd ratio after smoothing on the time shaft; This smoothing ratio as said the 2nd ratio, is added the above input signal in the signal after said the 2nd squelch
To after related said the 1st squelch of said audio frame, add under the situation of the above input signal in the signal,
With the said the 1st or the 2nd ratio of the previous frame of this audio frame, said the 1st ratio as this audio frame adds the above input signal in the signal after said the 1st squelch.
10. Noise Suppression Device as claimed in claim 1 is characterized in that,
Said the 1st ratio is to obtain through following formula (A),
og=max(0,tg-G)…(A)
Here, og is the 1st ratio that should calculate,
Tg is that said target noise suppresses gain,
G is said noise suppression gain,
Max (a, b) expression function that bigger among a and b value is returned.
11. Noise Suppression Device as claimed in claim 2 is characterized in that,
Said the 2nd ratio is to obtain through following formula (A),
og=max(0,tg-G)…(A)
Here, og is the 2nd ratio that should calculate,
Tg is that said target noise suppresses gain,
G is said mean value gain,
Max (a, b) expression function that bigger among a and b value is returned.
12. a noise suppressing method is characterized in that, comprises:
Noise spectrum is inferred step, and it infers the noise spectrum that is included in this input signal based on input signal respectively to K frequency band, and wherein, K is the natural number more than or equal to 2;
The 1st gain calculation step, its infer based on said noise spectrum step noise spectrum infer the result, calculate said K frequency band noise suppression gain separately; And
Former sound additional step; It is in signal after the 1st squelch that after said input signal is used said noise suppression gain, obtains; Add the above input signal with the 1st ratio; Wherein, the 1st ratio is based on that this noise suppression gain and predetermined target noise suppress the difference of gain and definite.
13. noise suppressing method as claimed in claim 12 is characterized in that,
Also comprise the 2nd gain calculation step, it calculates the mean value gain relevant with a said K frequency band to said noise suppression gain,
In said former sound additional step; Using to said input signal after the 2nd squelch that obtains after the gain of said mean value in the signal; Add the above input signal with the 2nd ratio, wherein, the 2nd ratio is based on difference that the gain of this mean value and said target noise suppress to gain and definite.
14. noise suppressing method as claimed in claim 13 is characterized in that,
Also comprise the audio detection step, it is through to whether comprising audio frequency in the said input signal detecting with passing in time, thereby this input signal is divided into audio frame that comprises this audio frequency and the noise frame that does not comprise this audio frequency,
Signal is in said input signal, to be equivalent to obtain after the said mean value gain of the certain applications of said noise frame after said the 2nd squelch.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2008274858A JP5526524B2 (en) | 2008-10-24 | 2008-10-24 | Noise suppression device and noise suppression method |
JP2008-274858 | 2008-10-24 |
Publications (2)
Publication Number | Publication Date |
---|---|
CN101727911A CN101727911A (en) | 2010-06-09 |
CN101727911B true CN101727911B (en) | 2012-07-04 |
Family
ID=42273495
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2009102055351A Expired - Fee Related CN101727911B (en) | 2008-10-24 | 2009-10-26 | Noise suppression device and noise suppression method |
Country Status (3)
Country | Link |
---|---|
JP (1) | JP5526524B2 (en) |
KR (1) | KR101088558B1 (en) |
CN (1) | CN101727911B (en) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9173025B2 (en) | 2012-02-08 | 2015-10-27 | Dolby Laboratories Licensing Corporation | Combined suppression of noise, echo, and out-of-location signals |
US8712076B2 (en) | 2012-02-08 | 2014-04-29 | Dolby Laboratories Licensing Corporation | Post-processing including median filtering of noise suppression gains |
CN104915930B (en) * | 2014-03-14 | 2019-05-28 | 深圳迈瑞生物医疗电子股份有限公司 | The method and device of grey level compensation and noise suppressed is carried out to image |
WO2016038704A1 (en) * | 2014-09-10 | 2016-03-17 | 三菱電機株式会社 | Noise suppression apparatus, noise suppression method and noise suppression program |
JP6559576B2 (en) | 2016-01-05 | 2019-08-14 | 株式会社東芝 | Noise suppression device, noise suppression method, and program |
WO2017135484A1 (en) * | 2016-02-05 | 2017-08-10 | 볼보 컨스트럭션 이큅먼트 에이비 | Variable cooling-fan apparatus of construction machine |
EP3312838A1 (en) | 2016-10-18 | 2018-04-25 | Fraunhofer Gesellschaft zur Förderung der Angewand | Apparatus and method for processing an audio signal |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1430778A (en) * | 2001-03-28 | 2003-07-16 | 三菱电机株式会社 | Noise suppressor |
CN1624767A (en) * | 2003-12-03 | 2005-06-08 | 富士通株式会社 | Noise reduction apparatus and noise reducing method |
CN101727910A (en) * | 2008-10-24 | 2010-06-09 | 雅马哈株式会社 | Noise suppression device and method |
Family Cites Families (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3454402B2 (en) * | 1996-11-28 | 2003-10-06 | 日本電信電話株式会社 | Band division type noise reduction method |
JP3459363B2 (en) * | 1998-09-07 | 2003-10-20 | 日本電信電話株式会社 | Noise reduction processing method, device thereof, and program storage medium |
JP3454190B2 (en) * | 1999-06-09 | 2003-10-06 | 三菱電機株式会社 | Noise suppression apparatus and method |
JP2002140100A (en) * | 2000-11-02 | 2002-05-17 | Matsushita Electric Ind Co Ltd | Noise suppressing device |
JP2002169599A (en) * | 2000-11-30 | 2002-06-14 | Toshiba Corp | Noise suppressing method and electronic equipment |
JP4162604B2 (en) * | 2004-01-08 | 2008-10-08 | 株式会社東芝 | Noise suppression device and noise suppression method |
JP4518817B2 (en) * | 2004-03-09 | 2010-08-04 | 日本電信電話株式会社 | Sound collection method, sound collection device, and sound collection program |
US7957964B2 (en) | 2004-12-28 | 2011-06-07 | Pioneer Corporation | Apparatus and methods for noise suppression in sound signals |
KR100657948B1 (en) * | 2005-02-03 | 2006-12-14 | 삼성전자주식회사 | Speech enhancement apparatus and method |
JP4863713B2 (en) * | 2005-12-29 | 2012-01-25 | 富士通株式会社 | Noise suppression device, noise suppression method, and computer program |
US8275611B2 (en) | 2007-01-18 | 2012-09-25 | Stmicroelectronics Asia Pacific Pte., Ltd. | Adaptive noise suppression for digital speech signals |
-
2008
- 2008-10-24 JP JP2008274858A patent/JP5526524B2/en not_active Expired - Fee Related
-
2009
- 2009-10-23 KR KR1020090101092A patent/KR101088558B1/en not_active IP Right Cessation
- 2009-10-26 CN CN2009102055351A patent/CN101727911B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1430778A (en) * | 2001-03-28 | 2003-07-16 | 三菱电机株式会社 | Noise suppressor |
CN1624767A (en) * | 2003-12-03 | 2005-06-08 | 富士通株式会社 | Noise reduction apparatus and noise reducing method |
CN101727910A (en) * | 2008-10-24 | 2010-06-09 | 雅马哈株式会社 | Noise suppression device and method |
Non-Patent Citations (2)
Title |
---|
JP特开2005-257748A 2005.09.22 |
STANISLAV GRUDEN AND BALDOMIR ZAJC.Using Spectral Subtraction for Suppression of Noise in Speech Signals with Analog Integrated Circuits.《Analog Integrated Circuits and Signal Processing》.1999,第18卷(第2-3期), * |
Also Published As
Publication number | Publication date |
---|---|
CN101727911A (en) | 2010-06-09 |
KR20100045934A (en) | 2010-05-04 |
JP2010102201A (en) | 2010-05-06 |
KR101088558B1 (en) | 2011-12-05 |
JP5526524B2 (en) | 2014-06-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101727910B (en) | Noise suppression device and method | |
CN101727911B (en) | Noise suppression device and noise suppression method | |
Ma et al. | SNR loss: A new objective measure for predicting the intelligibility of noise-suppressed speech | |
EP2828852B1 (en) | Post-processing gains for signal enhancement | |
US8804977B2 (en) | Nonlinear reference signal processing for echo suppression | |
US20080140396A1 (en) | Model-based signal enhancement system | |
US8930184B2 (en) | Signal bandwidth extending apparatus | |
Yong et al. | Optimization and evaluation of sigmoid function with a priori SNR estimate for real-time speech enhancement | |
US8364479B2 (en) | System for speech signal enhancement in a noisy environment through corrective adjustment of spectral noise power density estimations | |
US10049678B2 (en) | System and method for suppressing transient noise in a multichannel system | |
JP5791092B2 (en) | Noise suppression method, apparatus, and program | |
CN101620855B (en) | Speech sound enhancement device | |
JP2015529847A (en) | Percentile filtering of noise reduction gain | |
US20080285773A1 (en) | Adaptive LPC noise reduction system | |
US20140177853A1 (en) | Sound processing device, sound processing method, and program | |
US20140149111A1 (en) | Speech enhancement apparatus and speech enhancement method | |
US20090122997A1 (en) | Audio processing apparatus and program | |
Xu et al. | Components loss for neural networks in mask-based speech enhancement | |
EP2490218B1 (en) | Method for interference suppression | |
US10332541B2 (en) | Determining noise and sound power level differences between primary and reference channels | |
Alam et al. | Perceptual improvement of Wiener filtering employing a post-filter | |
Bao et al. | A novel speech enhancement method using power spectra smooth in wiener filtering | |
Li et al. | Adaptive β-order generalized spectral subtraction for speech enhancement | |
KR20230120734A (en) | Voice enhancing method | |
CN102598128A (en) | Signal processing method, information processor, and signal processing program |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120704 Termination date: 20171026 |