WO2008106852A1 - A method and device for determining the classification of non-noise audio signal - Google Patents

Abstract

A method for determining the classification of non-noise audio signal includes: first extracting the spectral feature parameters of the non-noise audio signal; then determining the classification of the present non-noise audio signal in the scope of frequency domain according to the spectral feature parameters of said non-noise audio signal and the defined threshold values of the feature parameters.

Description

 Method and apparatus for determining a non-noise audio signal attribution category

TECHNICAL FIELD The present invention relates to the field of communications, and more particularly to techniques for determining a class of useful signal attribution. BACKGROUND OF THE INVENTION With the development of broadband technology, current audio signals are also diversified: not only speech, but also audio signals such as music, unvoiced sound, and various noises. The speech, music, and unvoiced audio signals are generally referred to as non-noise audio signals; various noise audio signals are referred to as noisy audio signals. In order to adopt a suitable codec algorithm for different audio signals, it is necessary to determine the attribution categories of various non-noise audio signals before encoding and decoding the non-noise audio signals.

 In the field of audio signal processing, there are currently some encoders capable of discriminating music signals and speech signals, such as AMR-WB (Adaptive Multi-Rate-Wideband) and SMV (Selectable Mode Vocoder). Coding standard). The basic idea of discriminating the music signal and the speech signal is as follows: Before encoding and decoding the audio signal, extracting the time domain characteristic parameter used in the encoding and decoding; then using the time domain characteristic parameter, the music signal in the audio signal is used And the voice signal is discriminated.

 It can be seen that the discriminating process of the above audio signal can only use the time domain characteristic parameters involved in the encoding algorithm. Therefore, the method for determining the attribution category of the audio signal must exist depending on the encoding algorithm, and has no independence and portability. . Summary of the invention

 Embodiments of the present invention provide a method and apparatus for determining a home category of a non-noise audio signal that can exist without the encoding algorithm.

 Embodiments of the present invention are implemented by the following technical solutions:

Embodiments of the present invention provide a method of determining a home category of a non-noise audio signal, including: Obtaining frequency characteristic parameters of the non-noise audio signal;

 In the frequency domain, the current non-noise audio signal attribution category is determined according to the spectral characteristic parameter of the non-noise audio signal and the set characteristic parameter value.

 An embodiment of the present invention further provides an apparatus for determining a belonging class of a non-noise audio signal, including: a feature parameter acquiring unit, configured to acquire a frequency characteristic parameter of the non-noise audio signal;

 The attribution category determining unit is configured to determine, in the frequency domain range, the current non-noise audio signal attribution category according to the spectral characteristic parameter of the non-noise audio signal and the set characteristic parameter threshold.

 An embodiment of the present invention further provides an unvoiced discriminating device, including:

 a first acquiring unit, configured to acquire a frequency characteristic parameter of the audio signal;

 The unvoiced discriminating unit is configured to perform a decision on the unvoiced attribution category of the current non-noise audio signal according to one or more of the acquired characteristic parameters and the corresponding characteristic parameter threshold: a time domain zero-crossing rate zcr; The energy ratio of the frequency band ratiol.

 An embodiment of the present invention further provides a voice discriminating device, including:

 a second acquiring unit, configured to acquire a spectral feature parameter of the audio signal;

 The speech discriminating unit is configured to perform a speech attribution category determination on the current non-noise audio signal according to one or more of the acquired characteristic parameters and the corresponding feature parameter reading: spectral fluctuation flux; spectral fluctuation variance var— Flux; spectral fluctuation variance moving average flux_var- mov; time domain zero-crossing rate zcr; x% spectral attenuation Rolloff-x.

 An embodiment of the present invention further provides a music discriminating device, including:

 a third acquiring unit, configured to acquire a frequency characteristic parameter of the audio signal;

 a music discriminating unit, configured to perform a music attribution category determination on the current non-noise audio signal according to one or more of the acquired characteristic parameters: and a corresponding feature parameter threshold: a spectral fluctuation variance moving average flux_var_mov; x%i Lu attenuates Rolloff x,

It can be seen from the specific implementation provided by the above embodiments of the present invention that the attribution type of the current non-noise audio signal is determined according to the spectral characteristic parameter of the non-noise audio signal, so that the embodiment of the present invention can be separated from the coding algorithm. It exists, thus being independent and portable. BRIEF DESCRIPTION OF DRAWINGS FIG. 1 is a flow chart of a first embodiment provided by the present invention;

 2 is a flow chart of initial decision logic in a first embodiment provided by the present invention;

 3 is a logic flow chart of a modification decision in the first embodiment provided by the present invention;

 Figure 4 is a schematic structural view of a second embodiment provided by the present invention;

 Figure 5 is a schematic structural view of a third embodiment provided by the present invention;

 Figure 6 is a schematic structural view of a fourth embodiment of the present invention;

 Figure 7 is a schematic structural view of a fifth embodiment provided by the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The first embodiment of the present invention provides a method for determining a home category of a non-noise audio signal. The implementation process is as shown in FIG. 1 and includes:

 Step S100: Acquire frequency characteristic parameters of the non-noise audio signal.

For the input non-noise audio signal, the frequency parameters it has include: short-term feature parameters and their class-time feature parameters. The short-term characteristic parameters include: spectral fluctuation, 95% spectral rolloff, x% 谙 attenuation Rolloff-x (eg, ⁵⁰ % spectral attenuation (Roiloff-half)), low frequency band occupying the full frequency band The energy ratio ratiol, the zero-crossing rate zcr (the zero crossing rate, zcr); the frequency domain zero-crossing rate fzcr; the length-time characteristic is the variance and moving average of each short-term characteristic parameter, such as the fluctuation variance flux_var; The volatility variance moving average flux—var—tnov; “the decay variance rolloff—var.

 In the first embodiment, the above characteristic parameters are counted by taking 10 frames, that is, the duration of 100 ms, and the definitions and calculation formulas of these characteristic parameters are given below:

Defining the i-th time-domain sample value representing a frame of the sound signal, where 0 ≤ <M; M represents the number of sample values of one frame signal; T represents the number of frames; u jw, is the signal spectrum of the ith frame; N is the FFT (Fast Fourier Transform, the length of the Fast Fourier Transform), flux is the ier wave of the i-th frame, and TT^ is the moving average of the ith frame spectrum, the moving average of the spectrum, and the moving average of the spectral attenuation. The following takes the sound signal with a sampling rate of 16 kHz as an example to describe the characteristic parameters in detail: 1. Spectral fluctuation flux and its derived 谙 fluctuation variance flux—var and i-wave variance variance moving average flux—var—mov.

The spectral fluctuation flux feature parameter describes the variation between frames and frames. For music signals, flux is relatively low and smooth, while the flux of speech signals is usually high and varies greatly. It can be calculated using Equation 1; the spectral fluctuation variance flux-var and the fluctuation variance moving average flux-var-mov are calculated using Equation 2 and Equation 3, respectively:

•Formula 1

 .Formula 2 1 i

Flux _ var _ mov = var flux _i =― ^ var Jlnx k)

 10

 Equation 3 where "οπ«(·) is a normalization function.

 2. The low frequency band accounts for the energy ratio ratio1 of the full frequency band.

 This characteristic parameter describes the ratio of the low-band sub-band energy to the total energy. Usually the mtiol of the voice signal is higher, and the mtiol of the music signal is lower. Its calculation formula is as shown in formula 4:

 Formula 4

3, 95% spectral decay (Rolloff), 50% i-attenuation (Rolloff-half) and spectral attenuation variance (rolloff-var).

Where Rolloff represents the position of the point occupying 95% of the energy; Rolloff half means 50% of the total band The location of the point of energy.

 Generally, the point at which the speech signal spectrum is attenuated is relatively low, and the point at which the spectrum of the music signal is attenuated is relatively high. The calculation formulas for RoUoff and rolloff-var are shown in Equation 5 and Equation 6, respectively:

Rolloffii) =max(∑[/ (k) <0.95*∑U jyw, (/) ) Equation 5 rolloff― var(z) =丄 (Rolloff(k)- RoUoff,)

 m k= -m

 Formula 6

The calculation formula for Rolloff-half is shown in Equation 7:

Rolloff Jialfii) = max(∑f/ _pw _i (k)≤0.5*∑U _pw, (I) ) Equation 7

4. Time domain zero-crossing rate zcr.

 This feature parameter is mainly used to detect unvoiced sound. Since the voice is unvoiced at intervals, a higher musical zcr will occur. Its calculation formula is as shown in Equation 8: zcr = ^ll{x(i)x(i-l)<0}

 ί

In Equation 8, the function Π{Α} indicates that when Α is _trut h, II{A} is 1; when A is f _a i _se , Π{Α} is 0.

5. Frequency domain zero-crossing rate fzcr.

 The fzcr represents a measure of the fluctuation of the energy of a frame signal at different frequencies in the frequency domain. For speech signals, fzcr can be seen as a preliminary algorithm for formants. It can be obtained by: intercepting at least one frequency signal of the non-noise audio signal frame; normalizing each of the intercepted frequency signals; and performing average removal on the normalized spectrum signal The finishing process, and calculate the zero-crossing rate of the collated spectrum signal. Specifically, it can be calculated using Equation 9 to Equation 13:

 1 W2(,)

U_ v _gi {t)= ∑ U_pwM

 N2(t) - Nl(t) «=wi

Formula S> For "e[Nl(i), N2( )], there is:

U― mov _l (t, n) = U _ movO _i (n)-U _ avg _t (t)

 Public ^ ao where ϋ - movOi (n) is as shown in Equation 11:

 U—movO, (n) = [U _ pw, n) + U __pw, (n-l) + U _ pw, (n + 1)]/ 3

 Formula 11 then has:

1 ^r_1

^( =—∑Il{O ^r _wo _J (r,w) [/_wOT(f,nl)<0} public fzcr{i) =∑K{t) Equation 13 where Ν1 and Ν2 are frequency domain With a starting point, for example, Ν1 = [188Hz, 1500Ηζ, 2500Ηζ, 3750Ηζ], Ν2=[1500Hz, 2500Hz, 3750Hz, 8000Hz]; the ί/_ is the signal spectrum of the i-th frame; the t/_ _wov (t, 0 is the moving average of the t sub-segments of the ith frame; and T represents the number of frames.

 After obtaining the above characteristic parameters, step S200 is performed to determine the current non-noise audio signal attribution category according to the characteristic parameters of the non-noise audio signal and the set characteristic parameter threshold in the frequency domain.

 In step S200, when the logic parameter determination is performed by using the combination of the above characteristic parameters, a preliminary logic determination is first performed, and the non-noise audio signal is initially classified into four categories: unvoiced, voice, music, and uncertain signal; and then the correction logic is performed. The judgment, that is, the uncertainty signal obtained after the preliminary logic determination is further judged, so that it can be attributed to voice or music. as follows:

 First, a preliminary logic decision is made to classify the non-noise audio signal into the initial classification of speech and music, which are divided into four categories: unvoiced, voice, music, and uncertain signals. The specific implementation process is shown in Figure 2:

In step S101, the voice flag and the music flag are set to 0, that is, Speech Jlag=0 and Music_flag=0. Next, make the following judgments at the same time: Step S102: Determine whether the current non-noise audio signal belongs to the unvoiced sound according to one or more of the following characteristic parameters: a time domain zero-crossing rate zcr, and a low frequency band occupying an energy ratio ratio1 of the full frequency band. as well as,

 Step S103, according to one or more of the following characteristic parameters: 谙 fluctuation flux; i undulation variance flux_var; spectral fluctuation variance moving average flux_var-mov; time domain zero-crossing rate zcr, determining whether the current non-noise audio signal belongs to voice. as well as,

 Step S104, determining whether the current non-noise audio signal belongs to the voice according to the x% spectrum attenuation Rolloff_X, such as the 50% spectral attenuation Rolloff-half characteristic parameter. as well as,

 Step S105, determining whether the current non-noise audio signal belongs according to the unvoiced trailing flag ZCR-hangover_flag, the spectral fluctuation trailing flag Flux-hangover_flag or the attenuating trailing flag Rollhalf_hangover_flag of the audio signal of the previous frame. For voice. as well as,

 Step S106, according to one or more of the following characteristic parameters: 诰 fluctuation variance moving average flux var mov; x% spectral attenuation Rolloff x, determining whether the current non-noise audio signal belongs to music. as well as,

 Step S107: Determine whether the current non-noise audio signal belongs to music according to the spectral fluctuation variance moving average tailing flag flux_var_mov_hangover_flag of the previous frame.

 In the above step S102, if it is determined that the current audio signal belongs to the unvoiced sound, step S108 is executed, that is, the unvoiced trailing flag ZCR_hangover_flag is set to a first set value, such as ZCR_hangover_flag=20. Then, step S109 is executed, that is, the unvoiced identifier is output. Otherwise, step S113 is executed, that is, the speech_flag=0 is maintained, indicating that the current non-noise frame is neither a voice class. The specific implementation can be carried out as follows:

 Determining whether one or more of the following conditions are satisfied: whether the time domain zero-crossing rate zcr is greater than the time-domain zero-crossing rate threshold THR_ZCR; whether the low-band-to-full-band energy ratio ratio1 is greater than the low-band-to-full-band energy ratio Threshold THR-RA; if one of the conditions satisfies the condition, it is determined that the current non-noise frame belongs to the unvoiced category, and the unvoiced trailing flag ZCR_hangover_flag is the first set value, such as ZCR_hangover_flag-20; otherwise, step S113 is performed. That is, keep Speech_flag=0.

In step S103, if it is determined that the current audio signal belongs to the voice, step S110 is performed, that is, The fluctuation trailing flag Flux-hangover_flag is the second set value, such as Flux-hangover_flag=20; then step S112 is executed to output the voice identifier, that is, set Speech_flag=l. Otherwise, step S113 is performed, that is, Speechjlag=0 is maintained, indicating that the current non-noise 桢 does not belong to the voice class. The specific implementation can be carried out as follows:

 Determine if one or more of the following conditions are met:

Whether the spectral fluctuation flu is greater than the consensus fluctuation threshold THR_FLUX; whether the spectral fluctuation variance flux-var is greater than the spectral fluctuation variance threshold THR-FLUX-VAR; whether the spectral fluctuation flux is greater than the first spectral fluctuation variance function 3⁄4 f, (flux_var), such as f^ Flux—var) = 0.7-20*flux—var; whether the spectral fluctuation flux is less than the second undulation variance function f ₂ (flux_var), such as f ₂ (flux_var)=8*(flux—var); whether zcr is greater than the spectrum The fluctuation variance moving average function f(flux-var-mov), such as f(flux_var_mov)= 60-2609* flux_var_mov; if one of the conditions is satisfied, it is determined that the current non-noise audio signal belongs to the speech class, and the spectrum is set. The fluctuation tailing flag Flux-hangover_flag is the second set value, such as Flux-hangover_flag=20; then set Spech_flag=l; otherwise, execute step SI 13, that is, keep the Speech_flag=0, indicating the current Non-noise frames are not part of the voice class.

 In step S104, if it is determined that the current audio signal belongs to the voice, step S111 is performed, that is, the spectral attenuation trailing flag Rollhalf_hangover_flag is a third set value, such as Rollhalf_hangover_flag=20; then step S112 is performed to output the voice identifier, that is, Speech_flag=L Otherwise, step S113 is performed, that is, keeping Speech_flag=0, indicating that the current non-noise frame does not belong to the voice class. The specific implementation can be carried out as follows:

 Determine if one or more of the following conditions are met:

 Whether the x% spectral attenuation Rolloff-half is less than the x% spectral attenuation threshold THR_ROLL; if so, it is determined that the current non-noise audio signal belongs to the speech class, and the spectral attenuation trailing flag Rollhalf_hangover_flag is the third set value, such as Rollhalf_hangover- Flag=20; Then set Speech_flag=l; otherwise, execute step S113, that is, keep Speech_flag=0, indicating that the current non-noise frame is a non-speech class.

In step S105, if it is determined that the current audio signal belongs to the voice, step S111 is performed, and the output is performed. Voice identification, ie, Speech_flag=l. Otherwise, step S113 is executed, that is, keeping Speech_flag=0, indicating that the current non-noise frame is a non-speech class. The specific implementation can be carried out as follows:

 Determine if one or more of the following conditions are met:

 Whether the unvoiced trailing flag ZCR_hangover_flag is greater than 0; 旙 fluctuation tailing flag Fluxjiangover—flag is greater than 0; and i is attenuated trailing flag Rollhalfjiangover—flag is greater than 0;

 If so, the current audio signal is considered to be at the voice, and then Speech_flag = l. Otherwise, it does not process, that is, keeps Speech_flag=0, indicating that the current non-noise frame is a non-speech class.

 In step S106, if it is determined that the current audio signal belongs to music, step S114 is performed, that is, the spectral fluctuation variance moving average trailing flag flux_var_mov_hangover_flag is a fourth set value, such as flux_var_mov_hangover_flag=20; Step S115, outputting a music identifier, that is, setting Music_flag=l. Otherwise, step S116 is executed, that is, music_flag is maintained, indicating that the current non-noise frame does not belong to the music class. The specific implementation can be carried out as follows:

 Determine if one or more of the following conditions are met:

The spectral fluctuation variance moving average flux — var — mov is less than the third x°/. Spectral decay function f ₃

(Rolloff_x), such as f ₃ (Rolloff - half) = 0.03-l/2400 * (Rolloff - half); spectral fluctuation variance moving average flux_var - mov is less than the fifth set value, such as the fifth set value = 0.005; Is the fluctuation variance moving average flux_var_mov smaller than the fourth x°/o spectral attenuation function f ₄ (Rolloff_x), such as f ₄ (Rolloff_hali = l/l 867*Rolloff_half-0.0486; spectral fluctuation variance moving average flux_ var_mov is less than spectral fluctuation Variance moving average threshold THR_FLUX_VAR_MOV;

 If one of the conditions is satisfied, it is determined that the current non-noise audio signal belongs to the music category, so the speech fluctuation variance moving average tailing flag flux_var_mov_hangover_flag is the fourth set value, such as flux_var_mov_hangover —flag=20. Then, Music_flag=l is set; otherwise, step S116 is executed, that is, Music_flag=0 is held, indicating that the current non-noise frame does not belong to the music class.

In step S107, it is determined whether the spectral fluctuation variance moving average tailing flag flux_var_mov_hangover_flag is greater than 0; if yes, the current audio signal is considered to belong to the sound Le, then set Music_flag=l. Otherwise, step SI16 is executed, that is, Music_flag=0 is maintained, indicating that the current non-noise frame does not belong to the music class.

 After the above implementation process, after the non-noise audio signal is judged, the following flag may be output:

Speech_flag = Music_flag = 1, Speech - flag = 0, and Music - flag = 0.

 Then, step S117 is performed, that is, according to the Speech_flag and Music_flag, the attribution category of the current non-noise audio signal is determined:

 When Speech_flag=l and Music_flag=l, it indicates that the current non-noise audio signal belongs to both speech and music; or when Speech_flag=0 and Music_flag=0, it indicates that the current non-noise audio signal is not attributed to speech. If it is not attributed to the music, then step S118 is performed, that is, the information that the non-noise audio signal belongs to the indeterminate signal UNCERTAIN is determined;

 When Speech_flag=l and Music_flag=0, it indicates that the non-noise audio signal belongs to the voice, and then step S119 is performed to determine that the non-noise audio signal belongs to the voice;

 When Speech_flag=0 and Music_flag=l, it indicates that the non-noise audio signal belongs to the music, and then step S120 is performed to determine that the current non-noise audio signal belongs to the music.

 Uncertain audio signals for decisions that are neither in the speech category nor in the music category

UNCERTAIN, further needs to determine the attribution class according to the audio environment before the audio signal. The specific judgment method is shown in Figure 3:

 Step S201, determining that the audio environment before the current non-noise audio signal is a voice audio environment, and is also a music environment;

 If the speech_continue_counter (continuous speech counter, the number of consecutively occurring speech audio signals before the current non-noise audio signal) >THR_SPEECH value is satisfied, the audio environment before the current non-noise audio signal is determined as Voice and audio environment;

 If Music_continue_counter (continuous music counter, indicating the number of consecutive music audio signals) >THR_MUSIC threshold is satisfied, it is determined that the audio environment before the current non-noise audio signal is a music audio environment.

If Speech— continue—count>THR—SPEECH 阚, or Music—continent—count>THR—The MUSIC threshold is not satisfied, indicating that the audio environment before the current non-noise audio signal is neither a voice environment nor a music environment. Then, step S205 is directly performed, that is, the non-noise audio signal is judged as an indeterminate audio signal.

 When it is determined that the audio environment before the current non-noise audio signal is a voice environment, step S202 is performed to determine, according to at least one of flux, flux_var, flux_var__mov, Rolloff_var, and fzcr of the current non-noise audio signal. Whether the current non-noise audio signal is attributed to the voice, if yes, executing step S204, that is, determining that the current non-noise audio signal is voice, and juxtaposing the voice signal flag Speech_flag=l; otherwise, performing step S205, determining the current non-noise audio signal For uncertain audio signals.

 The specific implementation process of step S202 is as follows:

 Determine if at least one of the following conditions is met: flux>THR—flux, flux—var>THR_flux_var, flux_var_mov>THR_flux_var_mov,

Rolloff_var>THR_Rolloff_var , fzcr< THR — fzcr;

 If one of the above conditions is satisfied, it is determined that the current non-noise audio signal is speech, and the speech signal flag Speech_flag = 1; otherwise, the current non-noise audio signal is determined to be an indeterminate audio signal.

 At this time, the thresholds THR-flux, THR_flux_var, and THR_flux_var-mov may be different from the corresponding thresholds set by the initial judgment process.

 When it is determined that the audio environment before the current non-noise audio signal is a music environment, step S203 is performed to determine whether the current non-noise audio signal is based on at least one of flux_var_mov, Rolloff_var, and fzcr of the current non-noise audio signal. Attributable to the music, if yes, determining that the current non-noise audio signal is music, and juxtaposing the music signal flag Music_flag=l; otherwise, performing step S205, that is, determining that the current non-noise audio signal is an indeterminate audio signal.

 The specific implementation process of step S204 is as follows:

 Determining whether at least one of the following conditions is satisfied: flux_var_mov<THR_flux_var-mov, Rolloff_var<THR_Rolloff_var, fzcr> THR-fzcr;

If one of the above conditions is satisfied, it is determined that the current non-noise audio signal is music, and the music signal is set. The flag Music_flag=l; otherwise, it determines that the current non-noise audio signal is an indeterminate audio signal.

 At this time, the threshold THR_flux_var_mov may be different from the corresponding threshold set by the initial judging process.

 For the audio signals determined by the steps S101 to S120 which belong to both the voice category and the music category, and the uncertain audio signals determined by the steps S201 to S205 that are neither the voice category nor the music category, further determination may be performed. The judgment, the method of judgment adopted is as follows:

 The uncertainty audio signal is subjected to a decision of a home class based on an audio signal preceding the current non-noise audio signal. details as follows:

 Determining, by the attribution category of the indeterminate audio signal, a attribution category of the audio signal immediately before the indeterminate audio signal; or determining, by the attribution category of the indeterminate audio signal, The category to which a relatively large proportion of the audio signal belongs.

 For the audio signals determined by the steps S101 to S120 that belong to both the voice category and the music category, and the uncertain audio signals determined by the steps S201 to S205 that are neither in the voice category nor in the music category, other The soft decision method, for the uncertain audio signal, performs the classification of the attribution class, for example, using the GMM (Gaussian mixture model) decision method for further classification.

 The above embodiment is described by taking the steps S101 to S107 simultaneously as an example, and the details of each step are the same as those of the above-mentioned implementation process, and are not specifically described herein.

 The second embodiment of the present invention is an apparatus for determining a home category of a non-noise audio signal, and has a structure as shown in FIG. 4, including: a feature parameter obtaining unit and a home category determining unit. The attribution class determining unit includes: a voiceless discriminating subunit, a voice discriminating subunit, and a music discriminating subunit, and the home class determining unit further includes: a determining subunit.

 The interaction of signals between the various units is as follows:

 The feature parameter obtaining unit acquires a feature parameter of the non-noise audio signal; the feature parameter includes at least one of the following:

Spectral fluctuation flux; spectral fluctuation variance flux_var; spectral fluctuation variance moving average flux_var_mov; The energy ratio of the frequency band to the full frequency band ratiol; 95% spectral attenuation Rolloff; x% spectral attenuation Rolloff-x, such as 50% spectral attenuation Rolloff_half; spectral attenuation variance rolloff-var; spectral amplitude variance magvar; time domain zero-crossing rate zcr; The frequency domain zero crossing rate fzcr.

 The attribution category determining unit determines, in the frequency domain range, the current non-noise audio signal attribution category according to the characteristic parameter of the non-noise audio signal and the set characteristic parameter threshold. The specific treatment is as follows:

 The unvoiced discriminant subunit performs a decision on the unvoiced attribution category of the current non-noise audio signal according to one or more of the obtained characteristic parameters and the corresponding characteristic parameter threshold: a time domain zero-crossing rate zcr; a low frequency band occupying the full frequency band The specific energy ratio ratio is the same as the related description in the first embodiment, and will not be described in detail here. as well as,

 The speech discriminating subunit performs a speech attribution category determination on the current non-noise audio signal according to one or more of the acquired characteristic parameters and the corresponding characteristic parameter threshold: a spectral fluctuation flux; a variability variance var-flux; Spectral fluctuation variance moving average flux_var-mov; time domain zero-crossing rate zcr; x% spectral attenuation Rolloff-x, such as 50% spectral attenuation Rolloff-half; the specific processing is the same as the related description in the first embodiment, here No longer described in detail. as well as,

 The music discriminating subunit performs a music attribution category decision on the current non-noise audio signal according to one or more of the acquired characteristic parameters and the corresponding characteristic parameter threshold: spectral fluctuation variance moving average flux_var_mov; x The % spectrum decays Rolloff_x, such as 50% spectral decay Rolloff-half. The specific processing procedure is the same as that in the first embodiment, and will not be described in detail herein.

 When the unvoiced audio sub-unit, the voice decision sub-unit or the music decision sub-unit determines that the current non-noise audio signal belongs to neither the voice class nor the music class, the home class determining unit passes the Describe a decision subunit to determine whether a voice audio environment or a music audio environment exists before the current non-noise audio signal;

When there is a voice audio environment before the current non-noise audio signal, according to one or more of the following characteristic parameters obtained: i scent fluctuation flux; spectral fluctuation variance var-flux; spectral fluctuation variance moving average flux_var_mov; spectral attenuation variance rolloff — var; frequency domain zero-crossing rate fzcr, and corresponding characteristic parameters The threshold value is used to determine the voice attribution category for the current non-noise audio signal that is neither voice nor music. The specific processing procedure is the same as that in the first embodiment, and will not be described in detail herein.

 When there is a music audio environment in front of the current non-noise audio signal, one or more of the following characteristic parameters are obtained: spectral fluctuation variance moving average flux_var_mov; spectral attenuation variance rolloff_var; frequency domain zero-crossing rate fzcr, And corresponding feature parameter thresholds, and the current non-noise audio signal that is neither voice nor music belongs to the voice attribution category. The specific processing procedure is the same as that in the first embodiment, and will not be described in detail herein.

 And an audio signal determined by the unvoiced decision subunit, the voice decision subunit, or the music decision subunit, belonging to both the voice category and the music category, and when determined by the one of the decision subunits to be neither a voice category After the undetermined audio signal of the music category is not further determined, the audio signal may be further determined by the determining subunit, and the following determining method may be adopted: according to the audio signal before the current non-noise audio signal, Uncertain audio signals are subject to decision of attribution class. That is, the attribution category of the indeterminate audio signal is determined as the attribution category of the audio signal immediately before the indeterminate audio signal; or the attribution category of the indeterminate audio signal is determined as the uncertainty Among the audio signals in front of the audio signal, the category to which the larger proportion of the signal belongs.

 Other soft decision methods can also be used to determine the attribution class for the uncertain audio signal, for example, using the GMM (Gaussian Mixture Model) decision.

 The third embodiment provided by the present invention is an unvoiced discriminating device, and its structure is as shown in FIG. 5, and includes: a first acquiring unit and an unvoiced discriminating unit.

 The first obtaining unit acquires a feature parameter of the audio signal; the feature parameter includes a time domain zero-crossing rate zcr; and/or, the low frequency band occupies an energy ratio ratio1 of the full frequency band.

The unvoiced discriminating unit performs a decision on the unvoiced attribution category of the current non-noise audio signal according to one or more of the acquired characteristic parameters and the corresponding characteristic parameter threshold: a time domain zero-crossing rate zcr; a low frequency band The energy ratio of the frequency band is mtiol. The specific processing procedure is the same as that in the first embodiment, and will not be described in detail herein. The fourth embodiment of the present invention is a voice discriminating device, and the structure thereof is as shown in FIG. 6, and includes: a second acquiring unit and a voice discriminating unit;

 The second obtaining unit acquires a feature parameter of the audio signal; the feature parameter includes one or more of the following feature parameters:

 Spectral fluctuation flux; spectral fluctuation variance var_flux; spectral fluctuation variance moving average flux_var-mov; time domain zero-crossing rate zcr; x% spectral attenuation Rolloff_x, such as 50% attenuation Rolloff-half.

 The speech discriminating unit performs a speech attribution category determination on the current non-noise audio signal according to one or more of the acquired characteristic parameters and a corresponding characteristic parameter threshold: spectral fluctuation flux; language fluctuation variance var-flux ; spectral fluctuation variance moving average flux - var - mov; time domain zero crossing rate zcr; x% spectral attenuation Rolloff - X, such as 50% 谙 decay Rolloff - half. The specific processing procedure is the same as that in the first embodiment, and will not be described in detail herein.

 A fifth embodiment of the present invention is a music discriminating device, which has a structure as shown in FIG. 7, and includes: a third acquiring unit and a music discriminating unit.

 The third obtaining unit acquires a feature parameter of the audio signal; the feature parameter includes one or more of the following feature parameters:

Spectral fluctuation variance moving average flux var mov; x% please attenuate Rolloff x, such as 50% i瞽 decay Rolloff_half ₀

 The music discriminating unit performs a music attribution category decision on the current non-noise audio signal according to one or more acquired feature parameters and a corresponding feature parameter threshold: a spectral fluctuation variance moving average flux var-mov; X%i attenuates Rolloff x, such as 50% i ridge attenuation Rolloff half. The specific processing is identical to the related description in the first embodiment and will not be described in detail herein.

 It can be seen from the specific implementation provided by the above embodiments of the present invention that the attribution type of the current non-noise audio signal is determined according to the spectral characteristic parameter of the non-noise audio signal, so that the embodiment of the present invention can be separated from the coding algorithm. It exists, thus being independent and portable.

It is apparent that those skilled in the art can make various modifications and variations to the invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention are within the scope of the present invention The present invention is also intended to cover such modifications and variations within the scope of the equivalents.

Claims

Rights request A method for determining a belonging class of a non-noise audio signal, comprising: acquiring a spectral characteristic parameter of a non-noise audio signal;  In the frequency domain, the current non-noise audio signal attribution category is determined according to the frequency characteristic parameter of the non-noise audio signal and the set characteristic parameter threshold.  2. The method of claim 1 wherein the characteristic parameters comprise at least one of:  • i-fluctuation flux; spectral fluctuation variance flux—var; verb fluctuation variance moving average flux—var— mov; 氐band occupies the energy ratio of the full band ratiol; 95% spectral attenuation Rolloff; x% spectral attenuation Rolloff—x; time domain Zero crossing rate zcr»  3. The method of claim 2, wherein the feature parameter further comprises at least one of the following:  Spectral attenuation variance rolloff— var; frequency domain zero-crossing rate fzcr.  4. The method according to claim 3, wherein the frequency domain zero crossing rate fzcr is obtained by:  Intercepting at least one piece of the spectral signal of the non-noise audio signal;  The segmented frequency signal is normalized; and the normalized frequency-transmission signal is subjected to the removal of the average value, and the zero-crossing rate of the sequenced frequency signal is calculated.  The method according to claim 2 or 3, wherein the determining the current non-noise audio signal attribution category according to the spectral characteristic parameter and the set characteristic parameter threshold comprises:  Obtaining one or more of the following characteristic parameters: a time domain zero-crossing rate zcr; a low frequency band occupying an energy ratio ratio1 of the full frequency band, and a corresponding characteristic parameter threshold, and performing a decision on the unvoiced attribution category of the current non-noise audio signal; and / or, According to one or more of the following characteristic parameters obtained: spectral fluctuation flux; spectral fluctuation variance Var-flux; spectral fluctuation variance moving average flux_var_mov; time domain zero-crossing rate zcr, and corresponding characteristic parameter threshold, for the current non-noise audio signal to make a voice attribution category decision; and / or,  Determining the current non-noise audio signal according to the obtained x% spectral attenuation Rolloff_X characteristic parameter and the corresponding characteristic parameter threshold; and/or,  Determining the speech attribution category of the current non-noise audio signal according to the unvoiced trailing flag, the spectral fluctuation tailing flag, and the spectral attenuation trailing flag of the audio signal of the previous frame; and/or,  Obtaining a music attribution category for the current non-noise audio signal according to one or more of the following characteristic parameters obtained: a spectral fluctuation variance moving average flux_var_mov; x% spectral attenuation Rolloff_x, and a corresponding characteristic parameter threshold; and/or,  The average non-noise audio signal is subjected to a decision of the music attribution category based on the spectral fluctuation variance of the previous frame of the audio signal moving the average tailing flag.  The method of claim 5, wherein the process of determining the unvoiced audio category of the current non-noise audio signal comprises:  Determining whether one or more of the following conditions are satisfied: whether the time domain zero-crossing rate zcr is greater than the time-domain zero-crossing rate threshold THR_ZCR; whether the low-band-to-full-band energy ratio ratio1 is greater than the low-band-to-full-band energy ratio threshold THR —RA;  If one of the conditions satisfies the condition, it is determined that the current non-noise frame belongs to the unvoiced category, and the unvoiced trailing flag is set to the first set value; otherwise, it is determined that the current non-noise frame does not belong to the voice class.  7. The method according to claim 5, wherein the one or more of the following characteristic parameters are obtained: a fluctuation flux; a spectral fluctuation variance var-flux; a spectral fluctuation variance moving average flux-var — mov; the time domain zero-crossing rate zcr, and the corresponding characteristic parameter threshold, the process of determining the voice attribution category of the current non-noise audio signal, specifically:  Determine if one or more of the following conditions are met: Whether the language fluctuation flux is greater than the spectral fluctuation threshold THRJFLUX; whether the spectral fluctuation variance flux_var is greater than the spectral fluctuation variance threshold THR_FLUX_VAR; whether the spectral fluctuation flux is greater than the first spectral fluctuation variance function f^flux-var); Two-spectrum fluctuation variance function f ₂ (flux_var); whether zcr Greater than i volatility variance moving average function f(flux_var_mov);  If one of the conditions is satisfied, it is determined that the current non-noise audio signal belongs to the voice class, and the spectral fluctuation tail flag is the second set value; otherwise, it is determined that the current non-noise audio signal does not belong to the voice class.  The method according to claim 5, wherein the performing the voice attribution category on the current non-noise audio signal according to the acquired x%i-attenuated Rolloff_X feature parameter and the corresponding feature parameter threshold The process of the judgment includes:  Determining whether x%i attenuates Rolloff_X is less than x% spectral attenuation threshold THR_ROLL; if yes, determines that the current non-noise audio signal belongs to the speech category, and the speech attenuation trailing flag is the third set value; otherwise, It is determined that the current non-noise audio signal does not belong to the voice class.  The method according to claim 5, wherein the unvoiced trailing flag, the spectral fluctuation trailing flag, and the attenuating trailing flag of the audio signal of the previous frame are used to perform a voice attribution category on the current non-noise audio signal. The process of the judgment specifically includes:  Determining whether at least one of the following conditions is satisfied: whether the unvoiced trailing flag of the audio signal of the previous frame is greater than 0; whether the spectral fluctuation trailing flag of the previous frame of the audio signal is greater than 0; spectral attenuation of the previous frame of the audio signal Whether the flag is greater than 0;  If one of the conditions satisfies the condition, it is determined that the current non-noise audio signal belongs to the voice; if the above condition is not satisfied, it is determined that the current non-noise audio signal does not belong to the voice.  10. The method according to claim 5, wherein the one or more of the acquired characteristic parameters are: spectral fluctuation variance moving average flux_var_mov; x°/. The spectral attenuation Rolloff_x, and the corresponding characteristic parameter threshold, the process of determining the music attribution category of the current non-noise audio signal, specifically including:  Determine if one or more of the following conditions are met: Whether the volatility variance moving average flux_var_ mov is less than the third x% pan fading function β (Rolloff_x); the spectral volatility variance moving average flux _ var — mov is less than the fifth set value; the spectral fluctuation variance moving average flux_var — mov is less than Fourth x%i fading function f4 (Rolloff-x); spectral fluctuation variance moving average flux_var- mov is less than the spectral fluctuation variance moving average threshold THR—FLUX—VAR—MO V;  If one of the conditions is satisfied, it is determined that the current non-noise audio signal belongs to the music category, and the spectral fluctuation variance moving average trailing flag is the fourth set value; otherwise, it is determined that the current non-noise audio signal does not belong to the music class.  The method according to claim 5, wherein the moving the average trailing flag according to the spectral fluctuation variance of the audio signal of the previous frame, and determining the music attribution category of the current non-noise audio signal, further includes :  Determining whether the spectral fluctuation variance moving average trailing flag of the previous frame audio signal is greater than 0, and if so, determining that the current non-noise audio signal belongs to music; otherwise, determining that the current non-noise audio signal does not belong to music.  The method according to claim 5, wherein when it is determined that the current non-noise audio signal is neither attributed to the voice category nor belongs to the music category, the method further includes: Determining whether there is a voice audio environment or a music audio environment before the current non-noise audio signal; when there is a voice audio environment before the current non-noise audio signal, it is judged whether one or more of the following conditions are satisfied: spectral fluctuation flux, spectral fluctuation variance var — Whether one or more of flux, moving variance flux average flu_var_mov, 镨 attenuation variance rol ff — var characteristic parameters are greater than the corresponding characteristic parameter threshold; whether the frequency domain zero crossing rate f _zcr is less than the corresponding characteristic parameter threshold; If one of the conditions is satisfied, determining that the current non-noise audio signal belongs to the voice; otherwise determining that the current non-noise audio signal is not attributable to the voice;  When there is a music audio environment before the current non-noise audio signal, it is determined whether one or more of the following conditions are met: Spectral fluctuation variance moving average flux_var_mov, spectral attenuation variance rolloff- var one or more of the characteristic parameters Whether it is less than the corresponding characteristic parameter threshold; whether the frequency domain zero-crossing rate fzcr is greater than the corresponding characteristic parameter threshold, and if one of the conditions is satisfied, determining that the current non-noise audio signal belongs to music; otherwise, determining that the current non-noise audio signal is not Belongs to music. 13. The method according to claim 5, wherein when it is determined that the current non-noise audio signal belongs to both the voice category and the music category, or neither belongs to the voice category nor belongs to the music category, Also includes:  Using a Gaussian mixture model to determine the attribution category for audio indeterminate audio signals that are both speech and music, or that are both non-speech and non-music; or  Determining, by the attribution category of the indeterminate audio signal, a attribution category of the audio signal immediately before the indeterminate audio signal; or  The attribution category of the indeterminate audio signal is determined as a category to which a signal having a larger proportion of the audio signal before the indeterminate audio signal belongs.  The method according to claim 12, wherein when it is determined that the current non-noise audio signal belongs to neither the voice category nor the music category, the method further includes:  Using a Gaussian mixture model to determine the attribution category for an audio indeterminate audio signal that belongs to both speech and music, or that belongs to both non-speech and non-music; or  Determining, by the attribution category of the indeterminate audio signal, a attribution category of the audio signal immediately before the indeterminate audio signal; or  The attribution category of the indeterminate audio signal is determined as a category to which a signal having a larger proportion of the audio signal before the indeterminate audio signal belongs.  A device for determining a belonging class of a non-noise audio signal, comprising: a feature parameter acquiring unit, configured to acquire a spectral feature parameter of the non-noise audio signal;  The attribution class determining unit is configured to determine, in the frequency domain range, the current non-noise audio signal attribution category according to the spectral characteristic parameter of the non-noise audio signal and the set characteristic parameter threshold.  16. The apparatus of claim 15, wherein the spectral characteristic parameter comprises at least one of:  Fluctuation flux; i-wave fluctuation variance flux-var; spectral fluctuation variance moving average flux-var- mov; 4氐 band dominates the energy ratio of the full band ratiol; 95% spectral attenuation Rolloff; x°/.谙Attenuation Rolloff—x; Time domain zero-crossing rate zcr. 17. The apparatus according to claim 16, wherein the frequency characteristic parameter further comprises at least one of: attenuating variance rolloff_var; frequency domain zero crossing rate fzcr. 18. The apparatus according to claim 16 or 17, wherein the attribution category determining unit comprises:  The unvoiced discriminant subunit is configured to: according to the acquired one or more of the following characteristic parameters: a time domain zero-crossing rate zcr; a low frequency band occupying an energy ratio ratio1 of the full frequency band, and a corresponding characteristic parameter threshold, for the current non-noise audio signal Conducting a judgment on the category of unvoiced sound; and,  a speech discriminating subunit, configured to: according to one or more of the following characteristic parameters obtained: i-fluctuation flux; spectral fluctuation variance var_flux; spectral fluctuation variance moving average flux_var-mov; time-domain zero-crossing rate zcr; % transmits the decaying Rolloff_x, and the corresponding characteristic parameter threshold, and performs the decision of the voice attribution category on the current non-noise audio signal;  a music discriminating subunit, configured to perform one or more of the following characteristic parameters according to the following: a spectral fluctuation variance moving average flux_var_mov; x%谙attenuation Rolloff_x, and a corresponding characteristic parameter threshold, and music on the current non-noise audio signal Judgment of attribution category.  The device according to claim 18, wherein the attribution category determining unit further comprises:  a decision subunit, configured to determine whether a voice audio environment or a music audio environment exists before the current non-noise audio signal when it is determined that the current non-noise audio signal belongs to neither the voice category nor the music category;  When there is a speech audio environment before the current non-noise audio signal, according to one or more of the following characteristic parameters obtained: language fluctuation flux; i-wave fluctuation variance var_flux; spectral fluctuation variance moving average flux_var-mov; spectral attenuation The variance rolloff_var; the frequency domain zero-crossing rate fzcr, and the corresponding characteristic parameter threshold, the speech attribution category of the current non-noise audio signal;  When there is a music audio environment before the current non-noise audio signal, according to one or more of the acquired characteristic parameters: spectral fluctuation variance moving average flux_var_mov; spectral attenuation variance rolloff_var; frequency domain zero-crossing rate fzcr, and corresponding The feature parameter threshold is used to determine the voice attribution category for the current non-noise audio signal that is neither speech nor music. 20. An unvoiced discriminating device, comprising: a first acquiring unit, configured to acquire a frequency i ridge feature parameter of the audio signal;  The unvoiced discriminating unit is configured to perform one or more of the following characteristic parameters: a time domain zero-crossing rate zcr; a low frequency band occupying an energy ratio ratio1 of the full frequency band, and a corresponding characteristic parameter threshold, and performing current non-noise audio signals The judgment of the unvoiced category.  21. A speech discriminating device, comprising:  a second acquiring unit, configured to acquire a spectral feature parameter of the audio signal;  a speech discriminating unit, configured to: according to one or more of the following characteristic parameters obtained: spectral fluctuation flux; spectral fluctuation variance var-flux; spectral fluctuation variance moving average flux_var-mov; time domain zero-crossing rate zcr; x% spectrum The Rolloff_x is attenuated, and the corresponding characteristic parameter threshold is used to determine the voice attribution category of the current non-noise audio signal.  22. A music discriminating device, comprising:  a third acquiring unit, configured to acquire a spectral feature parameter of the audio signal;  a music discriminating unit, configured to perform music according to one or more of the following characteristic parameters: a spectral fluctuation variance moving average flux var mov; x% spectral attenuation Rolloff x, and a corresponding characteristic parameter threshold, and performing music on the current non-noise audio signal Judgment of attribution category.