CN105047201A - Broadband excitation signal synthesis method based on segmented expansion - Google Patents

Abstract

The invention provides a broadband excitation signal synthesis method based on segmented expansion. According to the method, two methods of narrow-band excitation spectrum translation and white noise excitation are combined, narrow-band excitation signals pass through a low-pass filter, first frequency-band excitation signals are outputted, after the translation of the first frequency-band excitation signal spectrum and high-pass filtering of the first frequency-band excitation signals are performed, second frequency-band excitation signals are obtained, white noise via energy adjustment is multiplexed by gain factors to obtain third frequency-band excitation signals, and finally, the first frequency-band excitation signals, the second frequency-band excitation signals, and the third frequency-band excitation signals are synthesized to form complete broadband excitation signals. According to the method, the harmonic structure of the excitation signals can be well reserved, too many artificial harmonic waves are not introduced, the operation complexity of the algorism is not high, the realization is easy, the performance of the conventional bandwidth expansion technology is improved, and the communication quality of narrow-band voice is improved.

Description

A kind of wideband excitation signal synthetic method based on segment extension

Technical field

The present invention relates to voice process technology field, more specifically, relate to a kind of wideband excitation signal synthetic method based on segment extension.

Background technology

Traditional telephone network, can only narrowband speech communication owing to being subject to the restriction of its transmission bandwidth, although narrowband speech can meet basic communication requirement, communication quality but reduces greatly.Due to economic cause, old telephone network can not realize real wideband transmit at short notice.The appearance of artificial speech bandwidth expansion technique, solve this difficult problem, this technology does some process at the receiving end of telephone network, some disappearance band component are added to the narrow band voice signal artificially received, receiving terminal is made to export the higher wideband speech signal of acoustical quality, after the linear prediction model proposition of voice, present most artificial speech bandwidth expansion technique is all realize based on this model (also referred to as source-filter model) design, as shown in Figure 1, the bandwidth expansion of voice is divided into independently two steps to carry out by this model: spectrum envelope expansion and pumping signal expansion.

Wideband excitation signal after expansion exports wideband speech signal eventually through composite filter (filter coefficient is the parameter calculating gained in spectrum envelope expansion process) synthesis.Therefore, pumping signal expansion is a requisite step in bandwidth expansion technique.At present, main pumping signal expansion algorithm mainly contains three kinds: narrowband excitation spectrum folds/spectrum translation, white-noise excitation and harmonic excitation.Although these three kinds of methods all comparative maturities at present, their shortcoming is: should retain suitable harmonic structure in the pumping signal of synthesis, can not introduce again too much artificial harmonic wave, because too much artificial harmonic wave can damage original harmonic structure; The computational complexity of algorithm can not be too high, otherwise this algorithm will be difficult to realize.

Summary of the invention

The invention provides a kind of wideband excitation signal synthetic method based on segment extension, the method computational complexity not high and be easy to realize.

In order to reach above-mentioned technique effect, technical scheme of the present invention is as follows:

Based on a wideband excitation signal synthetic method for segment extension, comprise the following steps:

S1: narrow band voice signal is successively passed through analysis filter after up-sampling, framing pre-service, obtains narrowband excitation signal u _n;

S2: by narrowband excitation signal u _nby the low-pass filter that cutoff frequency is α 1, export the first band excitation signal u _l, to the first band excitation signal u _lcarry out spectrum translation, be the Hi-pass filter of α 1 again by a cutoff frequency by the signal after spectrum translation, obtain the second band excitation signal u _m;

S3: a unit white noise generator is set and produces white noise signal u _f, calculate the first band excitation signal u _lvariance obtain gain factor adjustment white noise signal u _fenergy and be multiplied by gain factor then be the Hi-pass filter of α 2 by a cutoff frequency, obtain the 3rd band excitation signal u _h;

S4: by the first band excitation signal u _l, the second band excitation signal u _mwith the 3rd band excitation signal u _hmerge and obtain wideband excitation signal u.

Further, white noise signal u after energy adjusting in described step S3 _ffrequency band without limit for width, the Hi-pass filter being α 2 by a cutoff frequency carrys out the spectrum component of filtering white noise α less than 2, obtains the 3rd band excitation signal u that frequency band range is 6500Hz-8000Hz _h.

Preferably, to be the span of α 1 be cutoff frequency: 3000-3500HZ.

Preferably, to be the span of α 2 be cutoff frequency: 6000-6500HZ.

Further, the first band excitation signal u in described step S4 _l, the second band excitation signal u _mwith the 3rd band excitation signal u _hall be band limiting and the pumping signal of respective band component non-overlapping copies, three pumping signals carried out sum operation in time domain, obtains the wideband excitation signal u that a frequency band is complete.

Compared with prior art, the beneficial effect of technical solution of the present invention is:

Narrowband excitation is composed translation and white-noise excitation two kinds of methods combining by the present invention, narrowband excitation signal is passed through low-pass filter, export the first band excitation signal, the second band excitation signal is obtained by after the first band excitation signal spectrum translation and high-pass filtering, by using the white noise of energy adjusting and as the 3rd band excitation signal after being multiplied by gain factor, finally by the first band excitation signal, second band excitation signal and the 3rd band excitation signal synthesize complete wideband excitation signal, the method can retain the harmonic structure of pumping signal preferably and not introduce too much artificial harmonic wave, the computational complexity of algorithm itself not high and be easy to realize, improve the performance of existing bandwidth expansion technique, improve the communication quality of narrowband speech.

Accompanying drawing explanation

Fig. 1 is the bandwidth expanding method schematic diagram based on source filter model;

Fig. 2 is the pumping signal synthetic method schematic diagram based on segment extension;

Fig. 3 is the structural representation of spectrum shift method;

Fig. 4 is the oscillogram of voice, the comparison diagram of sound spectrograph of broadband voice, narrowband speech, the present invention's synthesis.

Embodiment

Accompanying drawing, only for exemplary illustration, can not be interpreted as the restriction to this patent;

In order to better the present embodiment is described, some parts of accompanying drawing have omission, zoom in or out, and do not represent the size of actual product;

To those skilled in the art, in accompanying drawing, some known features and explanation thereof may be omitted is understandable.

Below in conjunction with drawings and Examples, technical scheme of the present invention is described further.

Embodiment 1

As shown in Figure 2, a kind of wideband excitation signal synthetic method based on segment extension, comprises the following steps:

S1: narrow band voice signal is successively passed through analysis filter after up-sampling, framing pre-service, obtains narrowband excitation signal u _n;

S2: by narrowband excitation signal u _nby the low-pass filter that cutoff frequency is α 1, export the first band excitation signal u _l, to the first band excitation signal u _lcarry out spectrum translation, be the Hi-pass filter of α 1 again by a cutoff frequency by the signal after spectrum translation, obtain the second band excitation signal u _m; Wherein, to be the span of α 1 be cutoff frequency: 3000-3500HZ;

S3: a unit white noise generator is set and produces white noise signal u _f, calculate the first band excitation signal u _lvariance obtain gain factor adjustment white noise signal u _fenergy and be multiplied by gain factor then be the Hi-pass filter of α 2 by a cutoff frequency, obtain the 3rd band excitation signal u _h; The span of α 2 is: 6000-6500HZ;

S4: by the first band excitation signal u _l, the second band excitation signal u _mwith the 3rd band excitation signal u _hmerge and obtain wideband excitation signal u.

In the present embodiment, first narrow band voice signal carries out up-sampling (sampling rate is 8000Hz), the pre-service such as framing, doing up-sampling process to narrow band signal is in order to the frequency band range by increasing its signal, expand for follow-up band component, pretreated narrow band signal is by obtaining narrowband excitation signal after analysis filter, this filter coefficient is determined by the parameter of speech linear predictive analytical calculation gained, the input signal that narrowband excitation signal is expanded as pumping signal, it is the condition precedent realizing pumping signal expansion, narrowband excitation signal is after up-sampling, sample frequency becomes 16000Hz from original 8000Hz, this signal is by obtaining low band excitation signal after low-pass filter, the parameter of corresponding low-pass filter is as follows:

Sample frequency: 16000Hz

Cut-off frequecy of passband: 3400Hz pass band damping: 1dB

Stopband cutoff frequency: 3450Hz stopband attenuation: 60dB

The the first band excitation signal u obtained _lbe low band excitation signal, its frequency band range is 0-3400Hz.

Due to before acquisition narrowband excitation signal, narrow band signal is through up-sampling pre-service, so there will be unnecessary band component in the frequency spectrum of narrowband excitation signal, so the low-pass filter that must be 3400Hz by a cutoff frequency carrys out the outer unnecessary composition of filtering narrowband excitation signal low-frequency band, obtain the low band excitation signal u that frequency band range is 0-3400Hz _l; Low band excitation signal u _lthe second band excitation signal u is exported by spectrum shift operations module _m, u _mbe midband pumping signal, compose the specific implementation step of shift operations module as shown in Figure 3: lower band excitation u _lbe 3100Hz with a frequency, amplitude be 2 cosine function be multiplied after be the Hi-pass filter of 3400Hz by a cutoff frequency again, output band scope is the midband pumping signal u of 3400-6500Hz _m; From the modulating characteristic of Fourier transform, by signal times with an angular frequency for ω _mcosine function, frequency domain is equal to the frequency band of this signal has been carried out length is ω _mtranslation, spectrum translation time-domain expression respectively Ru shown in (1) formula:

${\hat{u}}_e(k;m)={\hat{u}}_n(k;m)\·2cos\left({\mathrm{\ω}}_{M}k\right)---\left(1\right)$

Wherein, represent the midband pumping signal u of expansion _mthe kth data point of m frame, similarly, represent low band excitation signal u _lthe kth data point of m frame, ω in the method _mbe chosen for 2 π 3100 (rad/s).

(1) frequency-domain expression corresponding to formula is as shown in (2) formula:

Φ _e(ω)＝Φ _n(ω-ω _M)+Φ _n(ω+ω _M)(2)

Wherein, Φ _e(ω) power spectrum of signal after expression spectrum translation, Φ _n(ω) power spectrum of low band excitation signal.

As can be seen from formula (2), the many Φ of the power spectrum of signal after spectrum translation _n(ω-ω _m) power spectrum composition.Therefore, the signal demand after spectrum translation carrys out this unnecessary composition of filtering by a Hi-pass filter, the band component needed for reservation, and the optimum configurations of this wave filter is as follows:

Sample frequency: 16000Hz

Stopband cutoff frequency: 3350Hz stopband attenuation: 60dB

Cut-off frequecy of passband: 3400Hz pass band damping: 1dB

The power spectrum of gained midband pumping signal is such as formula shown in (3):

${\widehat{\mathrm{\Φ}}}_e\left(\mathrm{\ω}\right)={\mathrm{\Φ}}_n(\mathrm{\ω}+{\mathrm{\ω}}_M)---\left(3\right)$

Wherein, represent the power spectrum of midband pumping signal.

Calculate low band excitation signal u _lvariance a unit white noise generator is set and produces white noise signal, will the variance of gained be calculated be multiplied by white noise signal as gain factor and be the Hi-pass filter of 6500Hz by a cutoff frequency, form the 3rd band excitation signal u of frequency band range 6500-8000Hz _h, u _hit is high band excitation signal.Low band excitation signal u _l, midband pumping signal u _mwith high band excitation signal u _hmerge, export complete wideband excitation signal; Calculate the standard deviation of low band excitation signal, for the energy adjusting of high frequency band unit white noise, computing formula is for shown in formula (4):

${\mathrm{\σ}}_{{\hat{u}}_n}=\sqrt{\frac{1}{N_k}\underset{k-1}{\overset{N_k}{\Σ}}{\left({\hat{u}}_n\right(k)-\mathrm{\μ})}^2}---\left(4\right)$

In formula, for a kth data point of low band excitation signal, μ is the average of low band excitation signal.

Produce unit white noise signal with random function, this signal obeys standardized normal distribution, and its probability density function is such as formula shown in (5):

$f\left(x\right)=\frac{1}{\sqrt{2\mathrm{\π}}}\exp (-\frac{x^2}{2})---\left(5\right)$

Growth factor be multiplied by unit white noise signal obtain energy adjusting after white noise signal, the frequency band due to white noise is without limit for width, and at present required be the white-noise excitation signal of high frequency band.Therefore, the band component that the white noise signal after energy adjusting will come beyond this signal higher frequency band of filtering by a Hi-pass filter.The optimum configurations of this wave filter is as follows:

Sample frequency: 16000Hz

Stopband cutoff frequency: 6450Hz stopband attenuation: 60dB

Cut-off frequecy of passband: 6500Hz pass band damping: 1dB

By the white noise signal after Hi-pass filter, be exactly required high band excitation signal, its frequency band range is 6500-8000Hz.

Low band excitation signal, midband pumping signal and high band excitation signal are carried out sum operation in time domain, obtain complete wideband excitation signal, shown in (6):

${\hat{u}}_w(k;m)={\hat{u}}_n(k;m)+{\hat{u}}_e(k;m)+{\hat{u}}_h(k;m)---\left(6\right)$

In formula, represent the estimated value of a m frame kth data point of wideband excitation signal, similarly, with represent low band excitation signal u respectively _l, midband pumping signal u _mwith high band excitation signal u _ha m frame kth data dot values, experimental result is as shown in Figure 4.

The corresponding same or analogous parts of same or analogous label;

Describe in accompanying drawing position relationship for only for exemplary illustration, the restriction to this patent can not be interpreted as;

Obviously, the above embodiment of the present invention is only for example of the present invention is clearly described, and is not the restriction to embodiments of the present invention.For those of ordinary skill in the field, can also make other changes in different forms on the basis of the above description.Here exhaustive without the need to also giving all embodiments.All any amendments done within the spirit and principles in the present invention, equivalent to replace and improvement etc., within the protection domain that all should be included in the claims in the present invention.

Claims (5)

1., based on a wideband excitation signal synthetic method for segment extension, it is characterized in that, comprise the following steps:

S1: narrow band voice signal is successively passed through analysis filter after up-sampling, framing pre-service, obtains narrowband excitation signal u _n;

S2: by narrowband excitation signal u _nby the low-pass filter that cutoff frequency is α 1, export the first band excitation signal u _l, to the first band excitation signal u _lcarry out spectrum translation, be the Hi-pass filter of α 1 again by a cutoff frequency by the signal after spectrum translation, obtain the second band excitation signal u _m;

S3: a unit white noise generator is set and produces white noise signal u _f, calculate the first band excitation signal u _lvariance obtain gain factor adjustment white noise signal u _fenergy and be multiplied by gain factor then be the Hi-pass filter of α 2 by a cutoff frequency, obtain the 3rd band excitation signal u _h;

S4: by the first band excitation signal u _l, the second band excitation signal u _mwith the 3rd band excitation signal u _hmerge and obtain wideband excitation signal u.

2. the wideband excitation signal synthetic method based on segment extension according to claim 1, is characterized in that, white noise signal u after energy adjusting in described step S3 _ffrequency band without limit for width, the Hi-pass filter being α 2 by a cutoff frequency carrys out the spectrum component of filtering white noise α less than 2, obtains the 3rd band excitation signal u that frequency band range is 6500Hz-8000Hz _h.

3. the wideband excitation signal synthetic method based on segment extension according to claim 1, is characterized in that, to be the span of α 1 be cutoff frequency: 3000-3500HZ.

4. the wideband excitation signal synthetic method based on segment extension according to any one of claim 1-2, is characterized in that, to be the span of α 2 be cutoff frequency: 6000-6500HZ.

5. the wideband excitation signal synthetic method based on segment extension according to claim 1, is characterized in that, the first band excitation signal u in described step S4 _l, the second band excitation signal u _mwith the 3rd band excitation signal u _hall be band limiting and the pumping signal of respective band component non-overlapping copies, three pumping signals carried out sum operation in time domain, obtains the wideband excitation signal u that a frequency band is complete.