CN100492495C - Apparatus and method for detecting noise - Google Patents

Abstract

The invention discloses a noise detecting device and method, receiving a current frame signal and add one-step length to a count value as receiving a frame signal each time, and as the count value and the preset constant meet a fixed proportion relation or meets the fixed proportion relation but the current frame signal is forecast as a voice signal, according to local least frequency spectrum energy of the previous frame signal and the frequency spectrum energy of the current frame signal, calculating the local least frequency spectrum energy of the current frame signal, and as they meet the fixed proportion relation and the current frame signal is forecasted as a voice signal, subtracting one-step length from the current count value; as they meet the fixed proportion relation but the current frame signal is not forecasted as a voice signal, according to the temporary least frequency spectrum energy of the previous frame signal and the frequency spectrum energy of the current frame signal, calculating the local least frequency spectrum energy of the current frame signal, and after this, according to the frequency spectrum energy and local least frequency spectrum energy of the current frame signal, judging whether the current frame signal is a pure noise, avoiding the voice being estimated as a noise and raising the noise detecting accuracy.

Description

A kind of noise detection apparatus and method

Technical field

The present invention relates to the signal analysis technology field, be specifically related to a kind of noise detection apparatus and method.

Background technology

At present, in mobile terminal communication, noise is the subject matter that causes the call voice quality to reduce: the microphone of portable terminal is also gone into the ground unrest collection communication module of portable terminal when recording user speech.Because user's environment of living in is complicated, variation, thereby ambient noise signal is normally unsettled, its spectral characteristic also is protean, in addition, compare with voice signal, the energy of ambient noise signal also has than big-difference with environment is different, as: there is less ground unrest usually in office, and there is very strong ground unrest usually in subway station etc.

At present, before communication module sent in voice, it is carried out noise reduction process usually, reach the purpose that improves voice quality by sound enhancement method.Be applicable to that the speech enhancement technique in the mobile terminal communication must have following characteristic: time delay is very little, so that do not disturb normal communication; Can adapt to the variation of ground unrest, effectively suppress noise; Avoid introducing noise that artificial treatment brings as music noise etc., do not damage voice quality.In addition, this speech enhancement technique must satisfy the needs that move in digital signal processing (DSP) chip or other special chip.Walkaway is a requisite part in the monophony speech enhancement technique.Walkaway, i.e. spectral characteristic value by estimated noise signal, as: spectrum energies etc., come whether detection signal is noise.By can distinguishing voice and the noise in the signal to the estimation of spectral characteristics of noise value, thereby can follow the tracks of the ground unrest of continuous variation, reach the purpose of the noise in the erasure signal.

The flow process of carrying out walkaway in the prior art as shown in Figure 1, its key step is as follows:

Step 101: to the spectrum energy Y[i of present frame input signal on Frequency point i, n] carry out smoothly obtaining the smooth spectrum energy S of present frame input signal on Frequency point i _f[i, n].

Smoothing formula is: $s_f[i,n]=\underset{j=-w}{\overset{w}{\mathrm{\Σ}}}b[i+j]{\left|Y[i+j,n]\right|}^2---\left(1\right)$

Wherein, w is a frequency values, (2w+1) length of expression Hanning window, b[i+j] be the Hanning window value on Frequency point (i+j), j is a frequency values, and n is an integer, the totalframes of expression current input signal, Y[i+j, n] for present frame promptly n frame input signal at the spectrum energy on the Frequency point (i+j) as the FFT coefficient on Frequency point (i+j).

Step 102: to S _f[i, n] carries out the time recursive operation, obtains the recurrence smooth spectrum energy S[i of present frame input signal on Frequency point i, n].

The time recurrence formula is: S[i, n]=α _sS[i, n-1]+(1-α _s) S _f[i, n] (2)

Wherein, S[i, n] be the recurrence smooth spectrum energy of former frame input signal on Frequency point i of present frame, α _sBe constant and satisfied: 0＜α _s＜1.

Step 103: the local minimum recurrence smooth spectrum energy S that calculates the present frame input signal _Min[i, n]:

$\left\{\begin{array}{c}{S}_{\min }[i,n]=\min \{S_{\min }[i,n-1],S[i,n]\}\\ S_{\mathrm{tmp}}[i,n]=\min \{S_{\mathrm{tmp}}[i,n-1],S[i,n]\}\end{array}\right.,$ When n is not the integral multiple of constant L (3)

$\left\{\begin{array}{c}{S}_{\min }[i,n]=\min \{S_{\mathrm{tmp}}[i,n-1],S[i,n]\}\\ S_{\mathrm{tmp}}[i,n]=S[i,n]\end{array}\right.,$ When n is the integral multiple of constant L (4)

Wherein, S _Min[i, n-1] be present frame former frame promptly: the local minimum recurrence smooth spectrum energy of (n-1) frame input signal on Frequency point i, S _Tmp[i, n-1] is the interim minimum recurrence smooth spectrum energy of former frame input signal on Frequency point i of present frame, S _Tmp[i, n] is the interim minimum recurrence smooth spectrum energy of present frame input signal on Frequency point i.

Step 104: judge $\frac{S[i,n]}{S_{\min }[i,n]}<\mathrm{\&delta;}$ Whether set up, if judge that the present frame input signal is pure noise, with seasonal walkaway factor I[i, n]=0; Otherwise, judge that the present frame input signal is not pure noise, promptly have voice on the present frame input signal, with seasonal I[i, n]=1.Wherein, δ is a constant.

After this, can calculate there are voice in the present frame input signal on Frequency point i probability according to following formula $\tilde{p}[i,n]:$

$\tilde{p}[i,n]={\mathrm{\&alpha;}}_p\tilde{p}[i,n-1]+(1-{\mathrm{\&alpha;}}_p)I[i,n]---\left(5\right)$

Wherein,

For the former frame input signal of present frame exists the probability of voice, α on Frequency point i _pBe constant and satisfied: 0＜α _p＜1.As can be seen, work as I[i, n]=0 o'clock, $\tilde{p}[i,n]=0.$

Afterwards, calculate the noise spectrum energy λ of present frame input signal on Frequency point i according to following formula _d[i, n]:

${\widetilde{\mathrm{\&alpha;}}}_d[i,n]={\mathrm{\&alpha;}}_d[i,n]+(1-{\mathrm{\&alpha;}}_d[i,n])\tilde{p}[i,n];$

${\mathrm{\&lambda;}}_d[i,n]={\widetilde{\mathrm{\&alpha;}}}_d[i,n]{\mathrm{\&lambda;}}_d[i,n-1]+(1-{\widetilde{\mathrm{\&alpha;}}}_d[i,n]){\left|Y[i,n]\right|}^2---\left(6\right)$

Wherein, λ _d[i, n-1] is the noise spectrum energy of former frame input signal on Frequency point i of present frame, α _d[i, n] is constant and satisfies: 0＜α _d[i, n]＜1.

By above-mentioned steps as can be known: at the local minimum recurrence smooth spectrum energy S that seeks the present frame input signal _MinWhen [i, n], be the whole of L at the totalframes n of input signal

During several times, be according to the S in the formula (4) _Tmp[i, n]=S[i, n] calculate the interim minimum recurrence smooth spectrum energy S of present frame input signal _Tmp[i, n].Suppose: m is integer and n=mL, and is set in (n-1) frame input signal and does not have voice, so, and according to formula (3) the local minimum recurrence smooth spectrum energy S of (n-1) frame input signal as can be known _Min[i, n-1]=min{S _Min[i, n-2], S[i, n-1] }, that is: S _Min[i, n-1]≤S[i, n-1] at this moment,, obtain the interim minimum recurrence smooth spectrum energy S of n frame input signal so according to formula (4) if in n frame input signal, there are the higher voice of energy _Tmp[i, n]=S[i, n], and S[i, n] S[i, and n-1], so clearly: S _Tmp[i, n]〉S[i, n-1] 〉=S _Min[i, n-1], after this if at n～(n+L) all have voice in the frame input signal, the interim minimum recurrence smooth spectrum energy that when the frame input signal of n～(n+L), calculates: S so _Tmp[i, n]～S _Tmp[i, n+L] all can be greater than S _Min[i, n-1], and when (n+L) frame, can obtain the local minimum recurrence smooth spectrum energy S of (n+L) frame input signal according to formula (4) _Min[i, n+L]=min{S _Tmp[i, n+L-1], S[i, n+L], and: S[i, n+L] S[i, and n-1], clearly: S _Min[i, n+L]〉S[i, n-1] S _Min[i, n-1], therefore, according to said process as can be known, noise detecting method of the prior art can cause the value of local minimum recurrence smooth spectrum energy bigger than normal, thereby causes some voice to be estimated as noise mistakenly, this noise detecting method is applied in the voice enhancing, can make that voice signal is deducted mistakenly from original input signal, thereby cause the voice on some time domain to weaken suddenly, or sound not nature.

Summary of the invention

In view of this, fundamental purpose of the present invention is to provide a kind of noise detection apparatus and method, is estimated as noise mistakenly to avoid voice, improves the precision of walkaway.

For achieving the above object, technical scheme of the present invention is achieved in that

A kind of noise detection apparatus, this device comprises:

The spectrum energy computing module is used for received signal, and the spectrum energy of the current frame signal that calculates is outputed to minimal frequency energy computing module and noise detection module;

The voice prediction judge module, be used for received signal, and receive whenever that on current Frequency point a frame signal adds a step-length with the count value of self module, when the current count value of self module and predetermined constant satisfy fixed proportion and concern, judge whether current frame signal is predicted as voice, if, the current count value of self module is subtracted a step-length, and to voice indicator signal of minimal frequency energy computing module output;

Minimal frequency energy computing module, the spectrum energy that is used for the current frame signal of received spectrum energy computing module output, and on current Frequency point, whenever receive a spectrum energy value, the count value of self module is added a step-length, when the current count value of self module and predetermined constant do not satisfy fixed proportion and concern, will be according to the local minimal frequency energy of former frame signal and the spectrum energy of current frame signal, the local minimal frequency energy of the current frame signal that obtains outputs to noise detection module; When the current count value of self module and predetermined constant satisfy the fixed proportion relation and confiscate the voice indicator signal of voice prediction judge module output, will be according to the interim minimal frequency energy of former frame signal and the spectrum energy of current frame signal, the local minimal frequency energy of the current frame signal that obtains outputs to noise detection module; When receiving the voice indicator signal of voice prediction judge module output, will be according to the local minimal frequency energy of former frame signal and the spectrum energy of current frame signal, the local minimal frequency energy of the current frame signal that obtains outputs to noise detection module, and the current count value of self module is subtracted a step-length;

Noise detection module is used for the local minimal frequency energy according to the current frame signal of the spectrum energy of the current frame signal of spectrum energy computing module output and the output of minimal frequency energy computing module, judges whether current frame signal is pure noise.

Described device comprises that further there is the probability calculation module in voice, is used for having probability according to the voice that the pure noise indication signal or the non-pure noise indication signal of noise detection module output are calculated current frame signal,

And described noise detection module is further used for, and when current frame signal is pure noise, pure noise indication signal is outputed to voice have the probability calculation module; When current frame signal is not pure noise, non-pure noise indication signal is outputed to voice have the probability calculation module.

Described voice exist the probability calculation module to be further used for, and exist probability to output to the voice prediction judge module voice of current frame signal,

Described voice prediction judge module is further used for, receiving and preserve voice exists the voice of the current frame signal of probability calculation module output to have probability, and when current count value and predetermined constant satisfy fixed proportion and concern, whether the voice of judging the former frame signal that self preserves exist probability greater than the constant of self preserving, if greater than, to voice indicator signal of minimal frequency energy computing module output.

Described device further comprises noise spectrum energy computing module, be used for existing the voice of the current frame signal of probability calculation module output to have the spectrum energy of the current frame signal of probability and the output of spectrum energy computing module, calculate the noise spectrum energy of current frame signal according to voice.

Described voice prediction judge module comprises: voice exist posterior probability computing module and judge module, wherein:

There is the posterior probability computing module in voice, be used for received signal, and receive whenever that on current Frequency point a frame signal adds a step-length with the count value of self module, when the current count value of self module and predetermined constant satisfy fixed proportion and concern, calculate the posterior probability that has voice on the former frame signal, and will exist the posterior probability of voice to output to judge module on this former frame signal, be used for after receiving the voice indicator signal of judge module output, the current count value of self module is subtracted a step-length;

Judge module, be used to judge that voice exist on the former frame signal of posterior probability computing module output exists the posterior probability of voice whether greater than the constant of self preserving, if greater than, there is posterior probability computing module output voice indicator signal to minimal frequency energy computing module and voice.

Described voice prediction judge module comprises: priori snr computation module and judge module, wherein:

Priori snr computation module, be used for received signal, and receive whenever that on current Frequency point a frame signal adds a step-length with the count value of self module, when the current count value of self module and predetermined constant satisfy fixed proportion and concern, calculate the priori signal to noise ratio (S/N ratio) of current frame signal, and the priori signal to noise ratio (S/N ratio) of frame signal outputs to judge module before will being somebody's turn to do, and is used for after the voice indicator signal of receiving judge module output, and the current count value of self module is subtracted a step-length;

Whether judge module, the priori signal to noise ratio (S/N ratio) of preceding frame signal that is used to judge the output of priori snr computation module greater than the constant of self preserving, if greater than, to minimal frequency energy computing module and priori snr computation module output voice indicator signal.

A kind of noise detection apparatus, this device comprises:

Counter module, be used for frame number counting to the signal on the current Frequency point, whenever receive a frame signal, count value is added a step-length, this frame signal is outputed to spectrum energy computing module and voice prediction judge module, count value is outputed to voice prediction judge module and minimal frequency energy computing module; And when receiving the look-at-me of voice prediction judge module output, current count value is subtracted a step-length;

The spectrum energy computing module is used for the spectrum energy of computing counter module output signal, and this spectrum energy is outputed to minimal frequency energy computing module and noise detection module;

The voice prediction judge module, be used for when the count value of counter module output and predetermined constant satisfy fixed proportion and concern, whether the signal of judging counter module output is predicted as voice, if, send a look-at-me to counter module, and to voice indicator signal of minimal frequency energy computing module output;

Minimal frequency energy computing module, be used for when the count value of counter module output and predetermined constant do not satisfy fixed proportion and concern, will be according to the local minimal frequency energy of former frame signal and the spectrum energy of current frame signal, the local minimal frequency energy of the current frame signal that obtains outputs to noise detection module; When the count value of counter module output satisfies the fixed proportion relation with predetermined constant and confiscates the voice indicator signal that the voice prediction judge module exports, will be according to the interim minimal frequency energy of former frame signal and the spectrum energy of current frame signal, the local minimal frequency energy of the current frame signal that obtains outputs to noise detection module; When receiving the voice indicator signal of voice prediction judge module output, will be according to the local minimal frequency energy of former frame signal and the spectrum energy of current frame signal, the local minimal frequency energy of the current frame signal that obtains outputs to noise detection module;

Noise detection module is used for judging according to the spectrum energy of spectrum energy computing module output and the local minimal frequency energy of minimal frequency energy computing module output whether the present frame input signal is pure noise.

Described device comprises that further there is the probability calculation module in voice, is used for having probability according to the voice that the pure noise indication signal or the non-pure noise indication signal of noise detection module output are calculated current frame signal,

And described noise detection module is further used for, and when current frame signal is pure noise, pure noise indication signal is outputed to voice have the probability calculation module; When current frame signal is not pure noise, non-pure noise indication signal is outputed to voice have the probability calculation module.

Described voice exist the probability calculation module to be further used for, and exist probability to output to the voice prediction judge module voice of current frame signal,

Described voice prediction judge module is further used for, receiving and preserve voice exists the voice of the current frame signal of probability calculation module output to have probability, and when current count value and predetermined constant satisfy fixed proportion and concern, whether the voice of judging the former frame signal that self preserves exist probability greater than the constant of self preserving, if greater than, to voice indicator signal of minimal frequency energy computing module output.

Described device further comprises noise spectrum energy computing module, be used for existing the voice of the current frame signal of probability calculation module output to have the spectrum energy of the current frame signal of probability and the output of spectrum energy computing module, calculate the noise spectrum energy of current frame signal according to voice.

Described voice prediction judge module comprises: voice exist posterior probability computing module and judge module, wherein:

There is the posterior probability computing module in voice, be used for received signal, and when the current count value of counter module output and predetermined constant satisfy fixed proportion and concern, there is the posterior probability of voice on the calculating former frame signal, and will exists the posterior probability of voice to output to judge module on this former frame signal;

Judge module is used to judge that voice exist the posterior probability that has voice on the former frame signal of posterior probability computing module output whether greater than the constant of self preserving, if greater than, to minimal frequency energy computing module output voice indicator signal.

Described voice prediction judge module comprises: priori snr computation module and judge module, wherein:

Priori snr computation module, be used for received signal, and when the current count value of counter module output and predetermined constant satisfy fixed proportion and concerns, the priori signal to noise ratio (S/N ratio) of calculating current frame signal, and the priori signal to noise ratio (S/N ratio) that will be somebody's turn to do preceding frame signal outputs to judge module;

Whether judge module, the priori signal to noise ratio (S/N ratio) of preceding frame signal that is used to judge the output of priori snr computation module greater than the constant of self preserving, if greater than, to minimal frequency energy computing module output voice indicator signal.

A kind of noise detecting method all carries out following steps to all signal frames on each Frequency point, and this method comprises:

A, receive current frame signal, count value is added a step-length, calculate and preserve the spectrum energy of current frame signal, and judge whether current count value satisfies fixed proportion with predetermined constant and concern, if not, execution in step B; Otherwise, judge whether current frame signal is predicted as voice, if be predicted as voice, current count value is subtracted a step-length and execution in step B; If be not predicted to be voice, execution in step C;

B, according to the local minimal frequency energy of former frame signal and the spectrum energy of current frame signal, calculate and preserve the local minimal frequency energy of current frame signal; According to the interim minimal frequency energy of former frame signal and the spectrum energy of current frame signal, calculate and preserve the interim minimal frequency energy of current frame signal, then execution in step D;

C, according to the interim minimal frequency spectrum energy of former frame signal and the energy of current frame signal, calculate and preserve the local minimal frequency energy of current frame signal; According to the spectrum energy calculating of current frame signal and the interim minimal frequency energy of preservation current frame signal, execution in step D then;

Whether the ratio of D, the spectrum energy of judging current frame signal and local minimal frequency energy is less than preset value, if judge that current frame signal is pure noise, otherwise the judgement current frame signal is not pure noise.

This method further comprises: there is posterior probability in default voice;

Steps A is described judges whether current frame signal is predicted as voice and is specially:

Calculate the posterior probability that has voice on the former frame signal, judge then whether described posterior probability exists posterior probability greater than described default voice, if judge that current frame signal is predicted as voice; Otherwise, judge that current frame signal is not predicted to be voice.

This method further comprises: there is the priori signal to noise ratio (S/N ratio) in default voice;

Steps A is described judges whether current frame signal is predicted as voice and is specially:

Calculate the priori signal to noise ratio (S/N ratio) that has voice on the current frame signal, judge then whether described priori signal to noise ratio (S/N ratio) exists the priori signal to noise ratio (S/N ratio) greater than described default voice, if judge that current frame signal is predicted as voice; Otherwise, judge that current frame signal is not predicted to be voice.

This method further comprises: there is probability in default voice;

Steps A is described judges whether current frame signal is predicted as voice and is specially:

There is the probability of voice on the calculating former frame signal, and judges whether the probability that has voice on this former frame signal exists probability greater than described default voice, if judge that current frame signal is predicted as voice; Otherwise, judge that current frame signal is not predicted to be voice.

Described step B is specially:

The relatively local minimal frequency energy of former frame signal and the spectrum energy of current frame signal are got wherein little person as the local minimal frequency energy of current frame signal, and are preserved the local minimal frequency energy of this current frame signal; The relatively interim minimal frequency energy of former frame signal and the spectrum energy of current frame signal are got wherein little person as the interim minimal frequency energy of current frame signal, and are preserved the interim minimal frequency energy of this current frame signal, execution in step D then.

Described step C is specially:

The relatively interim minimal frequency energy of former frame signal and the spectrum energy of current frame signal are got wherein little person as the local minimal frequency energy of current frame signal, and are preserved the local minimal frequency energy of this current frame signal; With the spectrum energy of current frame signal interim minimal frequency energy, and preserve the interim minimal frequency energy of this current frame signal, execution in step D then as current frame signal.

The technical scheme that the present invention put down in writing has avoided voice to be estimated as noise mistakenly, has improved the precision of walkaway.The present invention is applied in the noise spectrum estimation and noise reduction process of signals with noise, the estimating noise spectrum is also eliminated noise effectively effectively.

Description of drawings

The process flow diagram of Fig. 1 for carrying out walkaway in the prior art;

Fig. 2 is the process flow diagram that carries out walkaway provided by the invention;

Fig. 3 is the device block diagram of the specific embodiment one of walkaway provided by the invention;

Fig. 4 is the device block diagram of the specific embodiment two of walkaway provided by the invention;

Fig. 5 is the composition frame chart one of voice prediction judge module provided by the invention;

Fig. 6 is the composition frame chart two of voice prediction judge module provided by the invention;

Fig. 7 is the device block diagram of the specific embodiment three of walkaway provided by the invention;

Fig. 8 is the device block diagram of the specific embodiment four of walkaway provided by the invention;

Fig. 9 is the composition frame chart three of voice prediction judge module provided by the invention;

Figure 10 is the composition frame chart four of voice prediction judge module provided by the invention.

Embodiment

The present invention is further described in more detail below in conjunction with drawings and the specific embodiments.

Fig. 2 is the process flow diagram that carries out walkaway provided by the invention, and as shown in Figure 2, its concrete steps are as follows:

Step 201: a counter is set on Frequency point i, whenever receives a frame input signal, count value adds 1, and establishing current count value is F[i], calculate and preserve the spectrum energy S[i of present frame input signal on Frequency point i, n].

Here, n is an integer, the totalframes of expression current input signal, S[i, n] can be according to the spectrum energy Y[i of present frame input signal on Frequency point i, n] and formula (1) and (2) calculate.

Step 202: judge F[i] whether be the integral multiple of predetermined constant L, if, execution in step 203; Otherwise, execution in step 204.

The value of L can rule of thumb determine, for example: when sampling rate is 8 kilobits/second, L=60 usually.

Step 203: the local minimal frequency energy S that calculates and preserve the present frame input signal according to formula (3) _Min[i, n] and interim minimal frequency energy S _Tmp[i, n], that is: relatively the former frame input signal of present frame on Frequency point i local minimal frequency energy and the energy of present frame input signal, get wherein smaller as the local minimal frequency energy S of present frame input signal _Min[i, n], simultaneously relatively the former frame input signal of present frame on Frequency point i interim minimal frequency energy and the spectrum energy of present frame input signal, get wherein smaller as the interim minimal frequency energy S of present frame input signal _Tmp[i, n], execution in step 207 then.

Step 204: judge whether the present frame input signal is predicted as voice, if, execution in step 205; Otherwise, execution in step 206.

Judging whether the present frame input signal is predicted as voice can be by one of following three kinds of mode:

Mode one: the former frame input signal that calculates present frame exists the posterior probability p[i of voice, n-1 on Frequency point i], judge p[i then, n-1] 〉=whether C1 sets up, if establishment judges that then the present frame input signal is predicted to be voice on Frequency point i; Otherwise, judge that the present frame input signal is not predicted to be voice on Frequency point i.Here, C1 is a constant, and satisfies: 0＜C1＜1, and occurrence can rule of thumb be determined; N-1 represents the former frame of present frame.

Mode two: calculate the priori signal to noise ratio (S/N ratio) ξ [i, n] of present frame input signal, judge then whether ξ [i, n] 〉=C2 sets up,, judge that then the present frame input signal is predicted to be voice on Frequency point i if set up; Otherwise, judge that the present frame input signal is not predicted to be voice on Frequency point i.Here, C2 is a constant, and generally satisfy: 0＜C2＜10, occurrence can rule of thumb be determined.

Mode three: there is the probability of voice in the former frame input signal that calculates present frame on Frequency point i

Judge $\tilde{p}[i,n-1]\&GreaterEqual;C3$ Whether set up,, judge that then the present frame input signal is predicted to be voice on Frequency point i if set up; Otherwise, judge that the present frame input signal is not predicted to be voice on Frequency point i.Here, C3 is a constant, and satisfies: 0＜C3＜1, occurrence can rule of thumb be determined.

Here,

Can obtain according to formula (5): $\tilde{p}[i,n-1]={\mathrm{\&alpha;}}_p\tilde{p}[i,n-2]+(1-{\mathrm{\&alpha;}}_p)I[i,n-1],$ Wherein, The probability that has voice on the front cross frame input signal for present frame, I[i, n-1] value as follows: $\frac{S[i,n-l]}{S_{\min }[i,n-l]}<\mathrm{\&delta;}$ During establishment, I[i, n-1]=0; $\frac{S[i,n-l]}{S_{\min }[i,n-l]}<\mathrm{\&delta;}$ When being false, I[i, n-1]=1, wherein, S[i, n-1] be the spectrum energy of the former frame input signal of present frame, S _Min[i, n-1] is the local minimal frequency energy of the former frame input signal of present frame.

Step 205: calculate and preserve S according to formula (3) _Min[i, n] and S _Tmp[i, n], and with (F[i]-1) renewal F[i], execution in step 207 then.

Step 206: calculate and preserve S according to formula (4) _Min[i, n] and S _Tmp[i, n], that is: relatively the former frame input signal of present frame on Frequency point i interim minimal frequency energy and the spectrum energy of present frame input signal, get wherein smaller as the local minimal frequency energy S of present frame input signal _Min[i, n] is simultaneously with the spectrum energy of the present frame input signal interim minimal frequency energy S as the present frame input signal _Tmp[i, n], execution in step 207 then.

Step 207: judge $\frac{S[i,n]}{S_{\min }[i,n]}<\mathrm{\&delta;}$ Whether set up, if judge that the present frame input signal is pure noise; Otherwise, judge that the present frame input signal is not pure noise.

The value of δ is determined by experience, satisfies usually: δ〉1.

If $\frac{S[i,n]}{S_{\min }[i,n]}<\mathrm{\&delta;}$ Set up I[i then, n]=0; If $\frac{S[i,n]}{S_{\min }[i,n]}<\mathrm{\&delta;}$ Be false I[i, n]=1, after this can calculate there are voice in the present frame input signal on Frequency point i probability according to formula (5)

Can calculate the noise spectrum energy λ of present frame input signal on Frequency point i according to formula (6) afterwards _d[i, n].

It is pointed out that in concrete the application, all will carry out the described processing in above step 201～207 each the frame input signal on the Frequency point i.

Fig. 3 is a noise detection apparatus block diagram provided by the invention, and as shown in Figure 3, it mainly comprises:

Spectrum energy computing module 31: be used for receiving inputted signal, and calculate the spectrum energy of present frame input signal on current Frequency point, and this spectrum energy is outputed to minimal frequency energy computing module 33 and noise detection module 34.

Voice prediction judge module 32: be used for receiving inputted signal, and on current Frequency point, set a counter, on current Frequency point, whenever receive a frame input signal count value of self counter is added 1, when the rolling counters forward value of current self module is the integral multiple of predetermined constant L, judge whether the present frame input signal is predicted as voice, if, the rolling counters forward value of current self module is subtracted one, and to voice indicator signal of minimal frequency energy computing module 33 outputs.

Minimal frequency energy computing module 33: the spectrum energy of present frame input signal on current Frequency point that is used for 31 outputs of received spectrum energy computing module, and on current Frequency point, set a counter, on current Frequency point, whenever receive a spectrum energy value count value of self counter is added 1, and when the rolling counters forward value of current self module is not the integral multiple of L, calculate the local minimal frequency energy of present frame input signal on current Frequency point and interim minimal frequency energy according to formula (3), and the local minimal frequency energy that will obtain outputs to noise detection module 34; In the rolling counters forward value of current self module is the integral multiple of L and when confiscating the voice indicator signal of voice prediction judge module 32 outputs, calculate the local minimal frequency energy of present frame input signal on current Frequency point and interim minimal frequency energy according to formula (4), and the local minimal frequency energy that will obtain outputs to noise detection module 34; When receiving the voice indicator signal of voice prediction judge module 32 outputs, calculate the local minimal frequency energy of present frame input signal on current Frequency point and interim minimal frequency energy according to formula (3), and the rolling counters forward value of current self module subtracted one, the local minimal frequency energy that will obtain simultaneously outputs to noise detection module 34.

Noise detection module 34: the spectrum energy of present frame input signal on current Frequency point that is used for 31 outputs of received spectrum energy computing module, and according to the local minimal frequency energy of this spectrum energy and minimal frequency energy computing module 33 outputs, judge whether the present frame input signal is pure noise on current Frequency point, and this judged result is outputed to the outside.

From the above mentioned as can be seen: the count value of the counter of voice prediction judge module 32 and minimal frequency energy computing module 33 is consistent.

Further, as shown in Figure 4, this device comprises that there is probability calculation module 35 in voice, is used for according to the signal value of noise detection module 34 outputs and the constant alpha of self preserving _p, calculate the voice of present frame input signal on current Frequency point and have probability, and exist probability to output to the outside these voice.

And, noise detection module 34 is further used for, when the present frame input signal is judged to be pure noise on current Frequency point, signal 0 is outputed to voice have probability calculation module 35, judge on current Frequency point at the present frame input signal and signal 1 to be outputed to voice have probability calculation module 35 when being not pure noise.

Further, as shown in Figure 4, this device comprises: noise spectrum energy computing module 36: be used for existing the present frame input signal of probability calculation module 35 outputs to have the spectrum energy of present frame input signal on current Frequency point of probability and 31 outputs of spectrum energy computing module at the voice on the current Frequency point according to voice, calculate the noise spectrum energy of present frame input signal on current Frequency point, and this noise spectrum energy is outputed to the outside.

Further, as shown in Figure 4, the voice of current frame signal on current Frequency point that voice exist probability calculation module 35 to be further used for obtaining exist probability to output to voice prediction judge module 32,

Simultaneously, voice prediction judge module 32 is further used for receiving and preserving voice and exists the voice of current frame signal on current Frequency point of probability calculation module 35 outputs to have probability, and when the rolling counters forward value of current self module is the integral multiple of L, whether the voice of former frame input signal on current Frequency point of judging the present frame that self preserves exist probability greater than the constant C 3 of self preserving, if greater than, to voice indicator signal of minimal frequency energy computing module 33 outputs.

Perhaps, further, as shown in Figure 5, voice prediction judge module 32 comprises: voice exist posterior probability computing module 3211 and judge module 3212, wherein:

There is posterior probability computing module 3211 in voice: be used for receiving inputted signal, and on current Frequency point, set a counter, on current Frequency point, whenever receive a frame input signal count value of self counter is added 1, when the rolling counters forward value of current self module is the integral multiple of L, there is the posterior probability of voice in the former frame input signal that calculates present frame on current Frequency point, and this posterior probability outputed to judge module 3212, after receiving the voice indicator signal that judge module 3212 is sent, the rolling counters forward value of current self module is subtracted one.

Voice exist the count value of the counter of the count value of counter of posterior probability computing module 3211 and voice prediction judge module 32, minimal frequency energy computing module 33 to be consistent.

Judge module 3212: be used to judge that voice exist the posterior probability of posterior probability computing module 3211 outputs whether greater than the constant C 1 of self preserving, if greater than, there is voice indicator signal of posterior probability computing module 3211 outputs to minimal frequency energy computing module 33 and voice.

Perhaps, further, as shown in Figure 6, voice judge module 32 comprises: priori snr computation module 3221 and judge module 3222, wherein:

Priori snr computation module 3221: be used for receiving inputted signal, and on current Frequency point, set a counter, on current Frequency point, whenever receive a frame input signal count value of self counter is added 1, when the rolling counters forward value of current self module is the integral multiple of L, calculate the priori signal to noise ratio (S/N ratio) of present frame input signal on current Frequency point, and this priori signal to noise ratio (S/N ratio) outputed to judge module 3222, after receiving the voice indicator signal that judge module 3222 is sent, the rolling counters forward value of current self module is subtracted one.

The count value of the counter of the count value of the counter of priori snr computation module 3221 and voice prediction judge module 32, minimal frequency energy computing module 33 is consistent.

Judge module 3222: be used to judge that whether the priori signal to noise ratio (S/N ratio) of present frame input signal on current Frequency point of priori snr computation module 3221 outputs be greater than the constant C 2 of self preserving, if greater than, to minimal frequency energy computing module 33 and voice indicator signal of priori snr computation module 3221 outputs.

Fig. 7 is the device block diagram of the specific embodiment three of walkaway provided by the invention, and as shown in Figure 7, it mainly comprises:

Counter module 40: be used for the frame number of the input signal on the current Frequency point is counted, whenever receive a frame input signal, count value is added 1, and this frame input signal outputed to spectrum energy computing module 41 and voice prediction judge module 42, count value is outputed to voice prediction judge module 42 and minimal frequency energy computing module 43; And when receiving the look-at-me of voice prediction judge module 42 outputs, current count value is subtracted 1.

Spectrum energy computing module 41: be used for the present frame input signal on the current Frequency point of count pick up device module 40 output, and calculate the spectrum energy of this present frame input signal on current Frequency point, and this spectrum energy is outputed to minimal frequency energy computing module 43 and noise detection module 44.

Voice prediction judge module 42: be used for the present frame input signal on the current Frequency point of count pick up device module 40 output, when the count value of counter module 40 outputs is the integral multiple of predetermined constant L, judge whether the present frame input signal is predicted as voice, if, send a look-at-me to counter module 40, and to voice indicator signal of minimal frequency energy computing module 43 outputs.

Minimal frequency energy computing module 43: the spectrum energy of present frame input signal on current Frequency point that is used for 41 outputs of received spectrum energy computing module, when the count value of counter module 40 outputs is not the integral multiple of L, calculate the local minimal frequency energy of present frame input signal on current Frequency point and interim minimal frequency energy according to formula (3), and the local minimal frequency energy that will obtain outputs to noise detection module 44; In the count value of counter module 40 output is the integral multiple of L and when confiscating the voice indicator signal of voice prediction judge module 42 outputs, calculate the local minimal frequency energy of present frame input signal on current Frequency point and interim minimal frequency energy according to formula (4), and the local minimal frequency energy that will obtain outputs to noise detection module 44; When receiving the voice indicator signal of voice prediction judge module 42 outputs, calculate the local minimal frequency energy of present frame input signal on current Frequency point and interim minimal frequency energy according to formula (3), the local minimal frequency energy that obtains is outputed to noise detection module 44.

Noise detection module 44: the spectrum energy of present frame input signal on current Frequency point that is used for 41 outputs of received spectrum energy computing module, and according to the local minimal frequency energy of this spectrum energy and minimal frequency energy computing module 43 outputs, judge whether the present frame input signal is pure noise on current Frequency point, and this judged result is outputed to the outside.

Fig. 8 is the device block diagram of the specific embodiment four of walkaway provided by the invention, and as shown in Figure 8, this device is compared with Fig. 7, further comprises:

There is probability calculation module 45 in voice, are used for according to the signal value of noise detection module 44 outputs and the constant alpha of self preserving _p, calculate the voice of present frame input signal on current Frequency point and have probability, and exist probability to output to the outside these voice.

And, noise detection module 44 is further used for, when the present frame input signal is judged to be pure noise on current Frequency point, signal 0 is outputed to voice have probability calculation module 45, judge on current Frequency point at the present frame input signal and signal 1 to be outputed to voice have probability calculation module 45 when being not pure noise.

As shown in Figure 8, this device further comprises: noise spectrum energy computing module 46: be used for existing the present frame input signal of probability calculation module 45 outputs to have the spectrum energy of present frame input signal on current Frequency point of probability and 41 outputs of spectrum energy computing module at the voice on the current Frequency point according to voice, calculate the noise spectrum energy of present frame input signal on current Frequency point, and this noise spectrum energy is outputed to the outside.

Further, as shown in Figure 8, the voice of current frame signal on current Frequency point that voice exist probability calculation module 45 to be further used for obtaining exist probability to output to voice prediction judge module 42,

Simultaneously, voice prediction judge module 42 is further used for receiving and preserving voice and exists the voice of current frame signal on current Frequency point of probability calculation module 45 outputs to have probability, and when the count value of counter module 40 outputs is the integral multiple of L, whether the voice of former frame input signal on current Frequency point of judging the present frame that self preserves exist probability greater than the constant C 3 of self preserving, if greater than, to voice indicator signal of minimal frequency energy computing module 43 outputs.

Perhaps, further, as shown in Figure 9, voice prediction judge module 42 comprises: voice exist posterior probability computing module 4211 and judge module 4212, wherein:

There is posterior probability computing module 4211 in voice: be used to receive and preserve the present frame input signal on the current Frequency point of counter module 40 outputs, and when the count value of counter module 40 outputs is the integral multiple of L, there is the posterior probability of voice in the former frame input signal that calculates present frame on current Frequency point, and this posterior probability is outputed to judge module 4212.

Judge module 4212: whether be used to judge the posterior probability that there is 4211 outputs of posterior probability computing module in voice greater than the constant C 1 of self preserving, if greater than, to voice indicator signal of minimal frequency energy computing module 43 outputs.

Perhaps, further, as shown in figure 10, voice prediction judge module 42 comprises: priori snr computation module 4221 and judge module 4222, wherein:

Priori snr computation module 4221: be used to receive and preserve the present frame input signal on the current Frequency point of counter module 40 outputs, and when the count value of counter module 40 outputs is the integral multiple of L, calculate the priori signal to noise ratio (S/N ratio) of present frame input signal on current Frequency point, and this priori signal to noise ratio (S/N ratio) is outputed to judge module 4222.

Judge module 4222: be used to judge that whether the priori signal to noise ratio (S/N ratio) of present frame input signal on current Frequency point of priori snr computation module 4221 outputs be greater than the constant C 2 of self preserving, if greater than, to voice indicator signal of minimal frequency energy computing module 43 outputs.

In the present invention, when whenever receiving a frame input signal on current Frequency point, count value can be added a default step-length, in above-mentioned specific embodiment, this step-length value is 1; In addition, in the present invention, judge count value F[i] whether satisfy the fixed proportion relation with predetermined constant L, in above-mentioned specific embodiment, this fixed proportion value is an integer.

In practice, noise detection apparatus provided by the invention and method, can be applicable to that signals with noise is estimated as the noise spectrum of: voice signal etc. and noise reduction process in.

The above only is process of the present invention and method embodiment, in order to restriction the present invention, all any modifications of being made within the spirit and principles in the present invention, is not equal to replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (18)

1, a kind of noise detection apparatus is characterized in that, this device comprises:

The spectrum energy computing module is used for received signal, and the spectrum energy of the current frame signal that calculates is outputed to minimal frequency energy computing module and noise detection module;

The voice prediction judge module, be used for received signal, and receive whenever that on current Frequency point a frame signal adds a step-length with the count value of voice prediction judge module, when the current count value of voice prediction judge module and predetermined constant satisfy fixed proportion and concern, judge whether current frame signal is predicted as voice, if, the current count value of voice prediction judge module is subtracted a step-length, and to voice indicator signal of minimal frequency energy computing module output;

Minimal frequency energy computing module, the spectrum energy that is used for the current frame signal of received spectrum energy computing module output, and on current Frequency point, whenever receive a spectrum energy value, the count value of minimal frequency energy computing module is added a step-length, when the current count value of minimal frequency energy computing module and predetermined constant do not satisfy fixed proportion and concern, will be according to the local minimal frequency energy of former frame signal and the spectrum energy of current frame signal, the local minimal frequency energy of the current frame signal that obtains outputs to noise detection module; When the current count value of minimal frequency energy computing module and predetermined constant satisfy the fixed proportion relation and confiscate the voice indicator signal of voice prediction judge module output, will be according to the interim minimal frequency energy of former frame signal and the spectrum energy of current frame signal, the local minimal frequency energy of the current frame signal that obtains outputs to noise detection module; When receiving the voice indicator signal of voice prediction judge module output, will be according to the local minimal frequency energy of former frame signal and the spectrum energy of current frame signal, the local minimal frequency energy of the current frame signal that obtains outputs to noise detection module, and the current count value of minimal frequency energy computing module is subtracted a step-length;

Noise detection module is used for the local minimal frequency energy according to the current frame signal of the spectrum energy of the current frame signal of spectrum energy computing module output and the output of minimal frequency energy computing module, judges whether current frame signal is pure noise.

2, device as claimed in claim 1, it is characterized in that, described device comprises that further there is the probability calculation module in voice, is used for having probability according to the voice that the pure noise indication signal or the non-pure noise indication signal of noise detection module output are calculated current frame signal

And described noise detection module is further used for, and when current frame signal is pure noise, pure noise indication signal is outputed to voice have the probability calculation module; When current frame signal is not pure noise, non-pure noise indication signal is outputed to voice have the probability calculation module.

3, device as claimed in claim 2 is characterized in that, described voice exist the probability calculation module to be further used for, and exists probability to output to the voice prediction judge module voice of current frame signal,

Described voice prediction judge module is further used for, receiving and preserve voice exists the voice of the current frame signal of probability calculation module output to have probability, and when current count value and predetermined constant satisfy fixed proportion and concern, the constant whether voice of judging the former frame signal that the voice prediction judge module is preserved exist probability to preserve greater than the voice prediction judge module, if greater than, to voice indicator signal of minimal frequency energy computing module output.

4, as claim 2 or 3 described devices, it is characterized in that, described device further comprises noise spectrum energy computing module, be used for existing the voice of the current frame signal of probability calculation module output to have the spectrum energy of the current frame signal of probability and the output of spectrum energy computing module, calculate the noise spectrum energy of current frame signal according to voice.

5, device as claimed in claim 1 is characterized in that, described voice prediction judge module comprises: voice exist posterior probability computing module and judge module, wherein:

There is the posterior probability computing module in voice, be used for received signal, and receive whenever that on current Frequency point a frame signal exists the count value of posterior probability computing module to add a step-length voice, when voice exist the current count value of posterior probability computing module and predetermined constant to satisfy fixed proportion to concern, calculate the posterior probability that has voice on the former frame signal, and will exist the posterior probability of voice to output to judge module on this former frame signal, be used for after the voice indicator signal of receiving judge module output, existing the current count value of posterior probability computing module to subtract a step-length voice;

Judge module, be used to judge the constant that voice exist the posterior probability that has voice on the former frame signal of posterior probability computing module output whether to preserve greater than judge module, if greater than, there is posterior probability computing module output voice indicator signal to minimal frequency energy computing module and voice.

6, device as claimed in claim 1 is characterized in that, described voice prediction judge module comprises: priori snr computation module and judge module, wherein:

Priori snr computation module, be used for received signal, and receive whenever that on current Frequency point a frame signal adds a step-length with the count value of priori snr computation module, when the current count value of priori snr computation module and predetermined constant satisfy fixed proportion and concern, calculate the priori signal to noise ratio (S/N ratio) of current frame signal, and the priori signal to noise ratio (S/N ratio) of frame signal outputs to judge module before will being somebody's turn to do, be used for after receiving the voice indicator signal of judge module output, the current count value of priori snr computation module is subtracted a step-length;

Judge module is used to judge the constant whether the priori signal to noise ratio (S/N ratio) of the preceding frame signal of priori snr computation module output preserves greater than judge module, if greater than, to minimal frequency energy computing module and priori snr computation module output voice indicator signal.

7, a kind of noise detection apparatus is characterized in that, this device comprises:

Counter module, be used for frame number counting to the signal on the current Frequency point, whenever receive a frame signal, count value is added a step-length, this frame signal is outputed to spectrum energy computing module and voice prediction judge module, count value is outputed to voice prediction judge module and minimal frequency energy computing module; And when receiving the look-at-me of voice prediction judge module output, current count value is subtracted a step-length;

The spectrum energy computing module is used for the spectrum energy of computing counter module output signal, and this spectrum energy is outputed to minimal frequency energy computing module and noise detection module;

The voice prediction judge module, be used for when the count value of counter module output and predetermined constant satisfy fixed proportion and concern, whether the signal of judging counter module output is predicted as voice, if, send a look-at-me to counter module, and to voice indicator signal of minimal frequency energy computing module output;

Minimal frequency energy computing module, be used for when the count value of counter module output and predetermined constant do not satisfy fixed proportion and concern, will be according to the local minimal frequency energy of former frame signal and the spectrum energy of current frame signal, the local minimal frequency energy of the current frame signal that obtains outputs to noise detection module; When the count value of counter module output satisfies the fixed proportion relation with predetermined constant and confiscates the voice indicator signal that the voice prediction judge module exports, will be according to the interim minimal frequency energy of former frame signal and the spectrum energy of current frame signal, the local minimal frequency energy of the current frame signal that obtains outputs to noise detection module; When receiving the voice indicator signal of voice prediction judge module output, will be according to the local minimal frequency energy of former frame signal and the spectrum energy of current frame signal, the local minimal frequency energy of the current frame signal that obtains outputs to noise detection module;

Noise detection module is used for judging according to the spectrum energy of spectrum energy computing module output and the local minimal frequency energy of minimal frequency energy computing module output whether the present frame input signal is pure noise.

8, device as claimed in claim 7, it is characterized in that, described device comprises that further there is the probability calculation module in voice, is used for having probability according to the voice that the pure noise indication signal or the non-pure noise indication signal of noise detection module output are calculated current frame signal

And described noise detection module is further used for, and when current frame signal is pure noise, pure noise indication signal is outputed to voice have the probability calculation module; When current frame signal is not pure noise, non-pure noise indication signal is outputed to voice have the probability calculation module.

9, device as claimed in claim 8 is characterized in that, described voice exist the probability calculation module to be further used for, and exists probability to output to the voice prediction judge module voice of current frame signal,

Described voice prediction judge module is further used for, receiving and preserve voice exists the voice of the current frame signal of probability calculation module output to have probability, and when current count value and predetermined constant satisfy fixed proportion and concern, the constant whether voice of judging the former frame signal that the voice prediction judge module is preserved exist probability to preserve greater than the voice prediction judge module, if greater than, to voice indicator signal of minimal frequency energy computing module output.

10, install as claimed in claim 8 or 9, it is characterized in that, described device further comprises noise spectrum energy computing module, be used for existing the voice of the current frame signal of probability calculation module output to have the spectrum energy of the current frame signal of probability and the output of spectrum energy computing module, calculate the noise spectrum energy of current frame signal according to voice.

11, device as claimed in claim 7 is characterized in that, described voice prediction judge module comprises: voice exist posterior probability computing module and judge module, wherein:

There is the posterior probability computing module in voice, be used for received signal, and when the current count value of counter module output and predetermined constant satisfy fixed proportion and concern, there is the posterior probability of voice on the calculating former frame signal, and will exists the posterior probability of voice to output to judge module on this former frame signal;

Judge module is used to judge the constant that voice exist the posterior probability that has voice on the former frame signal of posterior probability computing module output whether to preserve greater than judge module, if greater than, to minimal frequency energy computing module output voice indicator signal.

12, device as claimed in claim 7 is characterized in that, described voice prediction judge module comprises: priori snr computation module and judge module, wherein:

Priori snr computation module, be used for received signal, and when the current count value of counter module output and predetermined constant satisfy fixed proportion and concerns, the priori signal to noise ratio (S/N ratio) of calculating current frame signal, and the priori signal to noise ratio (S/N ratio) that will be somebody's turn to do preceding frame signal outputs to judge module;

Judge module is used to judge the constant whether the priori signal to noise ratio (S/N ratio) of the preceding frame signal of priori snr computation module output preserves greater than judge module, if greater than, to minimal frequency energy computing module output voice indicator signal.

13, a kind of noise detecting method is characterized in that, all signal frames on each Frequency point are all carried out following steps, and this method comprises:

A, receive current frame signal, count value is added a step-length, calculate and preserve the spectrum energy of current frame signal, and judge whether current count value satisfies fixed proportion with predetermined constant and concern, if not, execution in step B; Otherwise, judge whether current frame signal is predicted as voice, if be predicted as voice, current count value is subtracted a step-length and execution in step B; If be not predicted to be voice, execution in step C;

B, according to the local minimal frequency energy of former frame signal and the spectrum energy of current frame signal, calculate and preserve the local minimal frequency energy of current frame signal; According to the interim minimal frequency energy of former frame signal and the spectrum energy of current frame signal, calculate and preserve the interim minimal frequency energy of current frame signal, then execution in step D;

C, according to the interim minimal frequency spectrum energy of former frame signal and the energy of current frame signal, calculate and preserve the local minimal frequency energy of current frame signal; According to the spectrum energy calculating of current frame signal and the interim minimal frequency energy of preservation current frame signal, execution in step D then;

Whether the ratio of D, the spectrum energy of judging current frame signal and local minimal frequency energy is less than preset value, if judge that current frame signal is pure noise, otherwise the judgement current frame signal is not pure noise.

14, method as claimed in claim 13 is characterized in that, this method further comprises: there is posterior probability in default voice;

Steps A is described judges whether current frame signal is predicted as voice and is specially:

Calculate the posterior probability that has voice on the former frame signal, judge then whether described posterior probability exists posterior probability greater than described default voice, if judge that current frame signal is predicted as voice; Otherwise, judge that current frame signal is not predicted to be voice.

15, method as claimed in claim 13 is characterized in that, this method further comprises: there is the priori signal to noise ratio (S/N ratio) in default voice;

Steps A is described judges whether current frame signal is predicted as voice and is specially:

Calculate the priori signal to noise ratio (S/N ratio) that has voice on the current frame signal, judge then whether described priori signal to noise ratio (S/N ratio) exists the priori signal to noise ratio (S/N ratio) greater than described default voice, if judge that current frame signal is predicted as voice; Otherwise, judge that current frame signal is not predicted to be voice.

16, method as claimed in claim 13 is characterized in that, this method further comprises: there is probability in default voice;

Steps A is described judges whether current frame signal is predicted as voice and is specially:

There is the probability of voice on the calculating former frame signal, and judges whether the probability that has voice on this former frame signal exists probability greater than described default voice, if judge that current frame signal is predicted as voice; Otherwise, judge that current frame signal is not predicted to be voice.

17, method as claimed in claim 13 is characterized in that, described step B is specially:

The relatively local minimal frequency energy of former frame signal and the spectrum energy of current frame signal are got wherein little person as the local minimal frequency energy of current frame signal, and are preserved the local minimal frequency energy of this current frame signal; The relatively interim minimal frequency energy of former frame signal and the spectrum energy of current frame signal are got wherein little person as the interim minimal frequency energy of current frame signal, and are preserved the interim minimal frequency energy of this current frame signal, execution in step D then.

18, method as claimed in claim 13 is characterized in that, described step C is specially:

The relatively interim minimal frequency energy of former frame signal and the spectrum energy of current frame signal are got wherein little person as the local minimal frequency energy of current frame signal, and are preserved the local minimal frequency energy of this current frame signal; With the spectrum energy of current frame signal interim minimal frequency energy, and preserve the interim minimal frequency energy of this current frame signal, execution in step D then as current frame signal.

Images