CN101458930B - Excitation signal generation in bandwidth spreading and signal reconstruction method and apparatus - Google Patents

Abstract

The invention discloses a method for generating an excitation signal in bandwidth expansion, which recombines a narrow-band low-frequency signal through spectrum folding to generate a required high-frequency excitation signal. The invention also provides a corresponding reconstruction method and device for medium and high frequency signals with extended bandwidth. The scheme of the present invention uses low-frequency signals to generate high-frequency signals, and based on the harmonic characteristics of the low-frequency and high-frequency spectra of the signals, it can better expand both voice and music signals, and the spectrum folding method adopted also ensures that the high and low frequencies are connected The continuity of the signal spectrum at the place; the experiment proves that the scheme of the present invention is suitable for extending the 7-14kHz ultra-wideband signal.

Description

Exciting signal generation and signal reconstruction method and device in bandwidth extension

technical field

The invention relates to the technical field of bandwidth expansion, in particular to a method for generating an excitation signal in bandwidth expansion, a method for reconstructing a high-frequency signal, and a corresponding device. The invention is particularly suitable for ultra-wideband bandwidth expansion.

Background technique

Bandwidth Extension (BWE: BandWidth Extension) technology is a technology that expands a signal with a narrower frequency range to a wider frequency range by selecting an appropriate parameter model, thereby improving the quality of the perceived audio signal.

Usually under the conditions of limited coding rate, such as mobile and network environments, based on the auditory characteristic that the human ear is more sensitive to low-frequency signals, in order to achieve better coding effects, most of the available bits are generally allocated to low-frequency signals. signal, but because the high-frequency components still play an important role in the subjective impression of sound quality, it is still hoped that the high-frequency signal can be reconstructed as well as possible at the decoding end. The following is an introduction to a method (ITU-T, G.729.1) currently used to expand from narrowband speech (0 to 3.4kHz) to wideband speech (0 to 7kHz), which uses Time Domain Bandwidth Extension (TDBWE: Time Domain BandWidth Extension), the specific solutions include:

1. Encoding end

① Pretreatment

Perform spectrum folding on the input high-frequency signal obtained by sampling at a sampling rate of 16kHz, that is, fold the 4-8kHz frequency band of the high-frequency part of the input signal to the 0-4kHz part. This process is equivalent to folding the 160kHz frequency band of the high-frequency part Time-domain samples are multiplied by (-1) ⁿ . Then pass the folded signal through a 3/4 low-pass filter to filter out the 3-4kHz frequency band, which corresponds to the 7-8kHz part of the original frequency band, and the pre-processed signal is S'(n), n = 0, . . . , 159.

②Extraction of time-domain spectral envelope parameters

The 20ms S'(n) frame is subdivided into 16 segments with a length of 1.25ms, and each segment contains 10 samples. The time-domain spectrum envelope parameters are calculated for every 10 sample points, and the calculation formula is as follows:

${\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; env</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \{</span>\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 9</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; S</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 15</span>}<span\; class="notranslate">\; ,</span>$

A total of 16 time-domain spectral envelope parameters T _env (i) are obtained.

③ Extraction of spectrum envelope parameters in frequency domain

In G.729.1, the encoder only extracts the frequency domain parameters of the last 10ms subframe (80 samples) of the 20ms frame, and the decoder obtains the frequency domain parameters of the first 10ms subframe through interpolation. When calculating the frequency domain parameters, a 128-point Hanning window w _F is added to the sequence of the 10ms subframe after the S'(n) frame, which consists of the first 72 points of the 144-point rising Hanning window and the 112-point falling Hanning window The last 56 points of the Ning window are composed of the 72nd sample point at the junction; 32 samples are viewed in front of the window, 16 samples are viewed behind the window, and 80 samples of the current subframe are added to form a total of 128 points. The signal after windowing is:

^Sw (n)=S'(n)· _wF (n+31), n=-31, . . . , 96.

S ^w (n) is transformed from the time domain to the frequency domain by using Fast Fourier Transform (FFT: Fast Fourier Transform). The length of the FFT transformation is 64 to obtain S ^fft (n), n=0, . . . , 64. Due to the 3/4 low-pass filtering in the preprocessing process, only the first 3/4 of the spectral data is valid after being transformed into the frequency domain; and due to the symmetry of the FFT transformation, the first 32 frequency domain data It is only necessary to select the first 24 data to express the frequency band of 0-3kHz. According to the first 20 frequency domain data, the frequency domain spectrum envelope parameters are calculated as:

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; env</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \{</span>\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; j</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span>{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; S</span>}}^{\mathrm{<span\; class="notranslate">\; fft</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 11</span>}<span\; class="notranslate">\; ,</span>$

Where W _F (n) is a weighting function, W _F (0)=W _F (2)=0.5, W _F (1)=1.

④Quantification of parameters

The 16 T _env (i) and 12 F _env (j) are subjected to de-mean splitting vector quantization. First calculate the average value M _T of T _env (i), and quantize M _T with a 5bit scalar in the logarithmic domain; respectively calculate the residuals of T _env (i) and F _env (j) and the quantization scalar; then the 16 time domain The residual is split into two 8-dimensional vectors, and the same codebook is used to quantize them with 7bit respectively. The 12 frequency-domain residuals are split into three 4-dimensional vectors, and different codebooks are used. The first two 4-dimensional vectors are quantized with 5bit respectively. The last 4-dimensional vector is quantized with 4bit.

2. Decoder

① Incentive Generation

The bandwidth-extended excitation signal (Excitation Signal) is reconstructed from the decoding parameters of the core layer. The following core layer decoding parameters are used to generate the excitation signal for bandwidth extension: integer pitch delay T0, fractional pitch delay frac; energy Ec contributed by the fixed codebook, energy Ep contributed by the adaptive codebook; fixed base layer in the core layer The gain g _c of the codebook excitation c(n), c(n), the gain g _p of the adaptive codebook excitation v(n), v(n); the enhancement layer in the core layer enhances the excitation c'(n), The gain g _enh of c'(n).

Calculate the excitation ratio of the adaptive codebook and the fixed codebook (including the enhancement layer codebook) for each frame by estimating the gain contributions of unvoiced and voiced sounds, and then multiply the respective excitations of unvoiced and voiced sounds by the gain to form the initial excitation signal, and then calculate the initial excitation signal The post-processing of the pitch delay is performed according to the pitch delay and other parameters to obtain the final excitation signal exc(n). exc(n) also needs to pass a 3/4 low-pass filter to limit the frequency range to 0-3kHz.

②Decoding of parameters

16 time-domain spectral envelope parameters T _env (i) and 12 frequency-domain spectral envelope parameters F _env (j) are decoded from the code stream, and the decoding process is the inverse process of the quantization encoding process at the encoding end.

③Time domain spectrum envelope shaping

Time-domain shaping is mainly to adjust the energy of the excitation signal. Calculate the time-domain spectrum envelope parameters of the excitation signal exc(n) according to the calculation method of T _env (i) _at the encoding end, and obtain 16 T' _env (i), and then subtract T' _env ( i) Get the energy difference between the two, so as to obtain the energy range gain that should be adjusted:

gain=2^[T _env (i)-T' _env (i)];

Then, the excitation signal exc(n) of 160 samples is multiplied by the corresponding gain to restore the time-domain adjusted signal S ^T (n).

④Frequency Domain Spectral Envelope Shaping

The decoded frequency domain parameter F _env (j) represents the last 10 ms of the 20 ms frame, and the frequency domain parameters of the first 10 ms frame can be obtained by interpolating the frequency domain parameters of the current frame and the previous 20 ms frame. The domain parameters are denoted as F _env,1 (j) and F _env,2 (j), respectively.

Then, similar to the processing method in the time domain, S ^T (n) performs frequency-domain parameter extraction according to the calculation method of the encoding end, and extracts once every 10ms, and calculates two sets of frequency-domain parameters, which are recorded as F' _{env, 1} (j) , F' _{env, 2} (j). By the difference between F _{env, 1} (j), F _{env, 2} (j) and F' _{env, 1} (j), F' _{env, 2} (j) respectively, the adjustment range G _{F of two subframes is obtained, 1} ( j), G _{F, 2} (j). Since the frequency domain calculation is carried out in divided frequency bands, a filter bank is used to adjust the spectrum envelope of the signal frequency band corresponding to each frequency domain parameter respectively. Obviously, there are 12 filters in total, and G _{F, 1} (j ), G _{F, 2} (j) weight the coefficients of the filter bank respectively, and then filter the front and rear 10ms subframes respectively to obtain the signal output _SHB (n) after frequency domain shaping.

⑤ Post-processing of BWE

Since some burrs may occur after the double adjustment of the time domain and the frequency domain, an adaptive amplitude compression function is used for post-processing to reduce the deviation of the envelope. The post-processing method is to process once every 80 sample points, and divide it into three sections, 6 sample points in the front section, 70 sample points in the middle section, and 4 sample points in the last section. The output after the post-processing envelope adjustment is (per The lines are pre-paragraph, break and post-paragraph in order):

Wherein, T _env (i) is a time-domain spectral envelope parameter corresponding to the currently adjusted sample point.

Since the 4-8kHz high-frequency signal is folded to 0-4kHz at the encoding end, spectrum folding should be performed again when restoring at the decoding end. The folding method is similar to the spectral folding method at the encoding end. Since the reconstructed output signal is 0-3kHz, the frequency domain coefficients of 3-4kHz can be filled with 0 and folded to obtain the high-frequency reconstruction signal of 4-7kHz.

In the process of proposing the present invention, the inventors found that the excitation generation at the decoding end of the above-mentioned bandwidth extension technology is generated using a method similar to the binary excitation generation method in the speech generation model, which is suitable for coding speech signals, while the coding effect for music-like signals is better. It is poor; and the experiment proves that under the above excitation generation method, if the bandwidth expansion technology is used for 7-14kHz ultra-wideband expansion, the noise is large and the coding effect is poor, indicating that this technology is not suitable for application in ultra-wideband expansion.

The invention provides a method for generating an excitation signal in bandwidth expansion and a corresponding high-frequency signal reconstruction method and device, which are suitable for high-frequency reconstruction of audio signals such as speech and music in broadband and ultra-wideband expansion.

A method for generating an excitation signal in bandwidth expansion, comprising: generating a first excitation signal exc(n) whose frequency range is 0～ _B0 , n=0,...,N-1; performing frequency spectrum on exc(n) Fold to generate the second excitation signal exc ^fold (n) with a frequency range of B ₀ to 2B ₀ ; perform synthesis filtering on exc (n) and exc ^fold (n), and output a third excitation signal with a frequency range of 0 to 2B ₀ exc _HB (m), m=0, . . . , 2N-1, the third excitation signal exc _HB (m) is used as a high-frequency excitation signal to perform high-frequency signal reconstruction.

A kind of rebuilding method of high-frequency signal in bandwidth extension, comprising: generate excitation signal exc _HB (m) according to the generating method of aforementioned excitation signal, m=0,..., 2N-1; Decoding obtains time-domain spectrum envelope parameter T _env (i) and frequency-domain spectral envelope parameters F _env (j), where i=0, ..., I-1, j = 0, ..., J-1; according to T _env (i) for exc The time-domain spectral envelope of _HB (m) is adjusted, and each T _env (i) is adjusted correspondingly to a section including A time-domain samples in exc _HB (m), A≤2N/I, generating the time-domain adjusted Signal S ^T (m); adjust the frequency domain spectrum envelope of S ^T (m) according to F _env (j), and each F _env (j) corresponds to adjust the bandwidth of B ₁ in the frequency domain of S ^T (m) A sub-band, B ₁ ≤ B ₂ /J, B ₂ is the frequency bandwidth of S ^T (m), and generates the reconstructed signal S ^F (m) adjusted in the frequency domain; performs spectrum folding on S ^F (m), and generates frequency The high-frequency reconstructed signal _SHB (m) in the range of 2B ₀ -2B ₀ +B ₂ .

A kind of generating device of excitation signal in the bandwidth expansion, comprises: core decoding module, is used for the first excitation signal exc(n) that output frequency range is 0～B ₀ , n=0,...,N-1; Spectrum The folding module is used to carry out frequency spectrum folding to exc(n), and the output frequency range is the second excitation signal exc ^fold (n) of B ₀ ~ 2B ₀ ; the synthesis filtering module is used to exc (n) and exc ^fold (n ) for synthesis filtering, the output frequency range is the third excitation signal exc _HB (m) of 0～ _2B0 , m=0,..., 2N-1, and the third excitation signal exc _HB (m) is used as The high-frequency excitation signal is used to reconstruct the high-frequency signal.

A reconstruction device for a bandwidth-extended medium-high frequency signal, comprising: an excitation signal generation unit, adopting the logic structure of the excitation signal generation device described in any one of claims 15 to 17, for generating the excitation signal exc _HB (m), m=0,...,2N-1; the decoding unit is used to decode and output the time-domain spectral envelope parameter T _env (i) and the frequency-domain spectral envelope parameter F _env (j), wherein i=0, .. ., I _{- 1} , j=0, _. (i) Corresponding adjustment exc _HB (m) includes a section of A time-domain sample points, A≤2N/I, output time-domain adjusted signal S ^T (m); frequency domain shaping unit, used for according to F _env (j) Adjust the frequency domain spectral envelope of ^ST (m), each F _env (j) corresponds to adjust a subband with a bandwidth of B ₁ in the frequency domain of ^ST (m), B ₁ ≤ _{B 2} / J, B ₂ is the frequency bandwidth of S ^T (m), the reconstructed signal S ^F (m) after output frequency domain adjustment; Spectrum folding unit, is used for carrying out spectrum folding to the S ^F (m) of input, and the generation frequency range is High-frequency reconstruction signal _SHB (m) of 2B ₀ ~ 2B ₀ +B ₂ .

The above technical solution adopts the method of resynthesizing the narrow-band low-frequency signal through spectrum folding to generate the required high-frequency excitation signal; because the low-frequency signal is used to generate the high-frequency signal, based on the harmonic characteristics of the low-frequency and high-frequency spectrum of the signal, it can be used for speech and music. The signals are well expanded, and the spectrum folding method adopted also ensures the continuity of the signal spectrum at the junction of high and low frequencies; experiments have proved that it is not only suitable for bandwidth expansion of 4-7kHz frequency band signals, but also suitable for 7-14kHz ultra- Broadband signals are extended.

Description of drawings

1 is a schematic diagram of the steps of a method for generating an excitation signal according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a logical structure of an excitation signal generating device according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of steps of a high-frequency signal reconstruction method according to an embodiment of the present invention;

Fig. 4 is a schematic diagram of a logic structure of a high-frequency signal reconstruction device according to an embodiment of the present invention.

Detailed ways

An embodiment of the present invention provides a method for generating an excitation signal in bandwidth expansion, which recombines a narrowband low-frequency signal through spectrum folding to generate a required high-frequency excitation signal. Embodiments of the present invention also provide a corresponding method for reconstructing high-frequency signals in bandwidth expansion, and a device for generating an excitation signal and a device for reconstructing high-frequency signals in bandwidth expansion. Each will be described in detail below.

With reference to Fig. 1, the generation method of excitation signal in the bandwidth expansion of the embodiment of the present invention mainly comprises steps:

A1. Generate a first excitation signal with a frequency range from 0 to B ₀ , where the first excitation signal is usually a narrowband excitation signal.

In this embodiment, the narrowband excitation signal exc(n), n=0, . exc(n) can be reconstructed based on the coding method of the core layer at the coding end by using Code Excited Linear Prediction (CELP: Code Excited Linear Prediction), for example, the excitation signal reconstruction method in the aforementioned background technology.

In order to simplify the processing process and reduce the computational complexity, a simple and effective exc(n) generation method based on CELP is provided in this embodiment, including:

① Decode the core code stream to obtain fixed codebook excitation and adaptive codebook excitation and their respective gains.

According to the encoding method of the core layer at the encoder, the fixed codebook excitation can be composed of two parts: the fixed codebook excitation c(n) of the base layer and the enhanced excitation c'(n) of the enhancement layer, and the corresponding gains are g _c and g _enh respectively.

② Obtain exc(n) by superimposing fixed codebook excitation and adaptive codebook excitation according to their respective gain weights.

In the case where the fixed codebook excitation consists of two parts, the formula for exc(n) is:

exc(n)＝g _p ·v(n)+g _c ·c(n)+g _enh ·c'(n)

Among them, v(n) is the adaptive codebook excitation, and g _p is the gain of v(n).

Usually the frequency range of exc(n) is 0-4kHz, and one frame is composed of 160 time-domain samples with a duration of 20ms, that is, B ₀ =4kHz, N=160.

A2. Perform spectrum folding on exc(n) to generate a second excitation signal with a frequency range of B ₀ to 2B ₀ ; corresponding to the narrow-band and low-frequency nature of exc(n), the second excitation signal can be regarded as a narrow-band high-frequency signal exc ^fold (n).

This process is equivalent to multiplying the N time-domain samples of exc(n) by (-1) ⁿ .

A3. Perform synthesis filtering on exc(n) and exc ^fold (n), and output a third excitation signal with a frequency range of 0 to 2B ₀ . The third excitation signal is a high-frequency excitation signal with extended bandwidth.

The so-called synthesis filtering is to combine exc(n) and exc ^fold (n) spectrums to obtain a high frequency excitation signal exc _HB (m) whose bandwidth is extended to 0-2B ₀ , m=0,...,2N-1. An optional synthesis method is:

A quadrature mirror filter (QMF: Quandrature Mirror Filter) is used to perform quadrature mirror synthesis filtering on exc(n) and exc ^fold (n).

In addition, according to the needs of practical applications, low-pass, high-pass or band-pass filtering can be further performed on exc _HB (m) with a frequency range of 0 to 2B ₀ to output exc _HB (m) in a partial frequency range. Based on the current audio signal coding requirements for the frequency range, generally the frequency range required for broadband signals is 0-7kHz, including the low-frequency part of 0-4kHz and the high-frequency part of 4-7kHz; the frequency range required for ultra-wideband signals is 0~14kHz, including the low frequency part of 0~8kHz and the high frequency part of 8~14kHz, it can be seen that the bandwidth of the high frequency part is usually 3/4 of the low frequency part, so in this case, it is also necessary to The high-frequency excitations are further processed, namely:

A4. Perform 3/4 low-pass filtering on exc _HB (m) with a frequency range of 0 to 2B ₀ , and output exc _HB (m) with a frequency range of 0 to 3B ₀ /2.

The high-frequency excitation signal with a frequency range of 0 to 3B ₀ /2 can be used to reconstruct a broadband or ultra-wideband high-frequency signal with a frequency range of 2B ₀ to 3.5B ₀ .

The device for generating the excitation signal in the bandwidth extension of the embodiment of the present invention for performing the method for generating the above excitation signal will be described below. Referring to FIG. 2, its basic logical structure includes:

The core decoding module 101 is used to output the first excitation signal exc(n) whose frequency range is 0～ _B0 , n=0, ..., N-1; the core decoding module 101 can adopt a processing module based on CELP, and the output exc(n) can be divided into two paths, which are provided to the spectrum folding module 102 and the synthesis filter module 103 respectively;

The spectrum folding module 102 is used to perform spectrum folding on exc(n), and output the second excitation signal exc ^fold (n) whose frequency range is B ₀ to 2B ₀ ;

Synthetic filtering module 103, for carrying out synthetic filtering to exc (n) and exc ^fold (n), output frequency range is the 3rd excitation signal exc _HB (m) of 0～ _2B0 , m=0,...,2N-1 ; The synthesis filter module 103 may use an orthogonal mirror synthesis filter.

In addition, based on the description of the frequency range requirements of the excitation signal in the aforementioned generation method, the excitation signal generating device of this embodiment may further include:

The 3/4 low _- pass filter 104 is used to input the exc _HB (m) whose frequency range is 0～2B ₀ , perform 3/4 low-pass filtering on it, and output the exc HB (m) whose frequency range is 0～3B ₀ /2 m).

In order to better understand the above-mentioned embodiment, an application in UWB bandwidth expansion is taken as an example below to illustrate the above-mentioned excitation signal generation process: first, a frame of excitation signal exc of 0-4kHz is extracted by the core layer based on CELP coding (n) (160 sample points); then fold to 4～8kHz frequency band by the mode of spectrum folding, generate the excitation signal exc ^fold (n) (160 sample points) of 4～8kHz frequency band; then through QMF synthesis filter, Synthesize exc(n) and exc ^fold (n) to the required full-band excitation exc ^qmf (m) (320 samples), at this time the signal bandwidth is 0 ~ 8kHz; then the full-band excitation signal exc ^qmf (m ) is filtered by a 3/4 low-pass filter to obtain an excitation signal exc _HB (m) (320 samples) of 0-6kHz.

In the above embodiments of the excitation signal generation method and device, the required high-frequency excitation signal is generated by resynthesizing the narrow-band low-frequency signal through spectrum folding; since the low-frequency signal is used to generate the high-frequency signal, based on the low-frequency and high-frequency spectrum of the signal. Harmonic characteristics can better expand both speech and music signals, and solve the problem that the binary excitation generation method in the similar speech generation model used in the existing time-domain bandwidth expansion has a relatively poor coding effect on music-like signals. In addition, the spectrum folding method adopted also ensures the continuity of the high- and low-frequency signal spectrum at the junction; the experiment proves that the above excitation signal generation scheme is not only suitable for bandwidth expansion of 4-7kHz frequency band signals, but also suitable for 7-14kHz ultra-wideband The signal is extended.

The reconstruction method of the mid-high frequency signal with bandwidth extension based on the above excitation signal generation method according to the embodiment of the present invention will be described below. Referring to Figure 3, it mainly includes steps:

B1. Generate a high-frequency excitation signal.

The _generation method of _the high-frequency excitation signal exc _HB (m), m= _{0 ,} .

B2. Decoding and obtaining time-domain spectrum envelope parameters and frequency-domain spectrum envelope parameters.

According to the decoding method corresponding to the encoding method of the encoder, the time-domain spectral envelope parameter T _env (i), i=0, ..., I-1, and the frequency-domain spectral envelope parameter F _env (j) are decoded from the code stream , j=0, . . . , J−1, the specific encoding and decoding methods are not limited in this embodiment. It should be noted that this decoding step does not have a strict logical sequence requirement in the whole reconstruction process, it can be performed synchronously or sequentially with other steps, and it is not necessarily required to decode T _env (i) and F _env (j) at the same time, It is sufficient as long as the decoding of the corresponding parameter has been performed before using it in the reconstruction process.

B3. Adjust the time domain spectrum envelope of exc _HB (m) according to T _env (i) to generate a time domain adjusted signal S ^T (m).

The time-domain spectral envelope adjustment process is executed corresponding to the extraction process of the time-domain spectral envelope parameters at the encoding end, and each T _env (i) corresponds to adjusting a section of exc _HB (m) including A time-domain samples, A≤2N /I, that is, the adjusted number of samples may be all or part of 2N samples. The corresponding relationship between each T _env (i) and the adjusted sample point is the same as the corresponding relationship during the extraction process at the encoding end. A specific adjustment method may adopt, for example, the time-domain spectrum envelope adjustment method in the aforementioned background art.

In order to provide a better adjustment effect, a time-domain spectrum envelope adjustment method is provided in this embodiment, including:

① Calculate the time-domain spectrum envelope parameter T' _env (i) of exc _HB (m) according to the method of calculating T _env (i) at the encoder end.

The so-called way of calculating T _env (i) at the encoding end is the process of extracting T _env (i) of the high-frequency signal _Shb (m) to be encoded at the encoding end. _Shb (m) is usually obtained by preprocessing the high-frequency part of the signal to be encoded by the encoding end: first fold the high-frequency signal obtained by frequency division after sampling to the low-frequency band, and then perform low-pass filtering according to the frequency range requirements of the encoding . An example of a calculation method of T' _env (i) is as follows:

Divide the 2N sample points of exc _HB (m) into I segments, each segment has A sample points, and calculate the log domain energy of each segment

${{\mathrm{<span\; class="notranslate">\; T</span>}}^{<span\; class="notranslate">\; \text{'}</span>}}_{\mathrm{<span\; class="notranslate">\; env</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \{</span>\underset{\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; exc</span>}}_{\mathrm{<span\; class="notranslate">\; HB</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; .</span>$

Usually, 10 sample points can be taken as a segment, that is, A=10, and the number of T' _env (i) at this time is I=N/5.

② Calculate the time-domain preliminary gain factor g _T (i) according to the energy difference between T _env (i) and T' _env (i).

An example of the calculation method of g _T (i) is as follows:

_gT (i)=2^[ _Tenv (i) _-T'env (i)],

Apparently, each g _T (i) corresponds to a segment in exc _HB (m) including A time-domain samples, and the corresponding relationship is the same as that between T' _env (i) and exc _HB (m).

③ Interpolate each g _T (i) to obtain A gain factors.

Various interpolation methods can be adopted as required to expand each g _T (i) into A gain factors g _{T, i} (a), a=0,..., A-1, for example, each g _{T, i} (a) is equal to g _T (i). In order to obtain a better time-domain adjustment effect, in the case of A=10, a smooth interpolation algorithm is provided in this embodiment to calculate g _T,i (a):

g _{T, i} (a)=w _T (a) g _T (i)+[1-w _T (a)] g ^last _{T, i} (a); where, w _T (a) is a window function, g ^last _{T, i} (a) is the gain factor of the sample point corresponding to exc _HB (m) in the previous frame. w _T (a) is specifically:

${\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; cos</span>}<span\; class="notranslate">\; [</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\frac{\mathrm{<span\; class="notranslate">\; \&pi;</span>}}{\mathrm{<span\; class="notranslate">\; 6</span>}}<span\; class="notranslate">\; ]</span><span\; class="notranslate">\; \}</span>& <span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 4</span>}\\ \mathrm{<span\; class="notranslate">\; 1</span>}& <span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 9</span>}\end{array}\right.$

The above interpolation algorithm can be understood as smoothing the first 5 g _{T, i} (a) using the corresponding g ^last _{T, i} (a) obtained by smooth interpolation in the previous frame, and smoothing the last 5 g _{T, i} (a ) then take the value of g _T (i).

④ According to g _{T, i} (a), adjust the gain of A×I sample points of exc _HB (m) to obtain S ^T (m).

The time-domain spectral envelope shaping of exc _HB (m) is obtained by simple multiplication of the adjusted sample point value with the corresponding gain factor g _T,i (a):

S ^T (m) = g _{T, i} (a) · exc _HB (m).

B4. Adjust the frequency-domain spectral envelope of S ^T (m) according to F _env (j) to generate a frequency-domain adjusted reconstruction signal S ^F (m).

Similar to the time-domain spectral envelope adjustment process, the frequency-domain spectral envelope adjustment process is also performed corresponding to the extraction process of the frequency-domain spectral envelope parameters at the encoder, and each F _env (i) corresponds to the adjustment of S ^T (m) in the frequency domain The bandwidth is a subband of B ₁ , B ₁ ≤ _{B 2} /J, and B ₂ is the frequency bandwidth of ^ST (m), that is, exc _HB (m). The corresponding relationship between each F _env (j) and the adjusted frequency band is the same as the corresponding relationship during the extraction process at the encoding end. A specific adjustment method may be, for example, the frequency domain spectrum envelope adjustment method in the aforementioned background art.

In order to reduce the computational complexity and improve the adjustment effect, a frequency domain spectrum envelope adjustment method is provided in this embodiment, including:

①According to the method of calculating F _env (j) at the encoding end, perform time-frequency ^{transformation} on S ^T (m) to generate frequency domain signal S ^F1 (m) and calculate the frequency domain spectrum envelope parameter F' _env ( j).

The so-called method of calculating F _env (j) at the encoding end is the process of extracting F _env (j) of the high-frequency signal _Shb (m) to be encoded at the encoding end. An example of the calculation method of F' _env (i) is as follows:

Windowing w _TDAC (k) for ^ST (m) and last frame ^{ST, last} (m) to obtain windowed signal S ^w (k), k=0,..., 4N-1, where,

S ^w (k) = w _TDAC (k) S ^{T, last} (k), k = 0, ..., 2N-1,

S ^w (k) = w _TDAC (k) S ^T (k-2N), k = 2N, ..., 4N-1;

Perform discrete cosine transform (DCT: Diserete Cosine Transform) on S ^w (k) to generate S ^F1 (m). The specific transformation method can be modified discrete cosine transform (MDCT: Modified DCT),

${\mathrm{<span\; class="notranslate">\; S</span>}}^{\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 4</span>}\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; S</span>}}^{\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; cos</span>}<span\; class="notranslate">\; [</span>\frac{\mathrm{<span\; class="notranslate">\; \&pi;</span>}}{\mathrm{<span\; class="notranslate">\; 8</span>}\mathrm{<span\; class="notranslate">\; N</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span><span\; class="notranslate">\; ;</span>$

Take the first D×J samples of S ^F1 (m) to calculate F' _env (j),

${{\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \text{'}</span>}}_{\mathrm{<span\; class="notranslate">\; env</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \{</span>\underset{\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; S</span>}}^{\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; .</span>$

Since the generation of exc _HB (m) may have performed a 3/4 low-pass filtering process that limits the frequency range, in this case only the data in the 0-3B ₀ /2 frequency band is valid, so the time-frequency After transformation, it is only necessary to extract the first 3/2N points of the 2N frequency domain samples for calculating F' _env (j), at this time D×J=3/2N.

Usually, 16 samples can be taken as a sub-band, that is, D=16, and the number of F' _env (j) at this time is J=3N/32. In addition, the used window function w _TDAC (k) can choose the following sine window:

w _TDAC (k)=sin[(k+0.5)π/4N].

②Calculate the preliminary gain factor g _F (j) in the frequency domain according to the energy difference between F _env (j) and F' _env (j), each g _F (j) corresponds to S ^F1 (m) including D A segment of frequency domain samples, D×J≤2N.

An example of the calculation method of g _F (j) is as follows:

g _F (i) = 2^[F _env (j)-F' _env (j)],

The corresponding relationship between each g _F (i) and S ^F1 (m) sub-band is the same as the corresponding relationship between F' _env (i) and S ^F1 (m) sub-band.

③ Interpolate each g _F (j) to obtain D gain factors g _{F, j} (d), d=0, ..., D-1.

For a specific interpolation method, reference may be made to the aforementioned interpolation method for the time-domain gain factor, and of course other interpolation methods may also be used, which will not be repeated here.

④ Adjust the gain of D×J samples of S ^F1 (m) according to g _{F, j} (d), and generate the adjusted frequency domain signal S ^F2 (m). Similar to the adjustment of the time-domain spectrum envelope, simply multiply the frequency-domain sample point value with the corresponding gain factor g _{F, j} (d):

S ^F2 (m) = g _{F, j} (d) · S ^F1 (m).

⑤ Carry out the inverse transformation of the time-frequency transformation on S ^F2 (m) to obtain S ^F (m).

For example, if the MDCT is used to transform to the frequency domain before the frequency domain adjustment, then the inverse MDCT (IMDCT) is used to transform to the time domain.

B5. Perform spectrum folding on S ^F (m) to generate a high-frequency reconstruction signal _SHB (m) with a frequency range of 2B ₀ to 2B ₀ +B ₂ .

Since high-frequency signals are folded to low-frequency bands at the encoding end, spectrum folding should be performed again when restoring at the decoding end. The folding method is similar to the spectral folding method when the high-frequency signal is preprocessed at the encoder end. If during the reconstruction process, low-pass filtering is performed on the excitation signal based on the frequency range requirements of the encoding, at this time, the frequency domain coefficients of the high-frequency part removed by filtering can be filled with 0 and then folded to obtain the final high-frequency reconstruction signal.

Furthermore, due to the double adjustment of time domain and frequency domain in the above signal reconstruction process, it is likely to cause glitches in the reconstructed signal. In order to eliminate these glitches, the time-frequency adjusted signal S ^F can be adjusted before spectrum folding (m) Post-processing, that is, adding the following steps before step B5:

B51. Perform envelope adjustment on S ^F (m) using envelope adjustment thresholds limit ₁ (i) and limit ₂ (i). The adjusted S ^F (m) is:

In the part of m=m ₁ ~m ₂ , if |S ^F,old (m)|<limit ₁ (i), then S ^F (m)=S ^F,old (m),

In the part of m=m ₂ +1～m ₃ , if limit ₁ (i)≤|S ^{F, old} (m)|≤limit ₂ (i), then S ^F (m)=[S ^{F, old} ( m)-limit ₁ (i)]/2+limit ₁ (i),

In the part of m=m ₃ +1～m ₄ , if |S ^{F, old} (m)|>limit ₂ (i), then S ^F (m)=[S ^{F, old} (m)-limit ₂ ( i)]/16+limit ₂ (i), where S ^{F, old} (m) is S ^F (m) before envelope adjustment; limit ₁ (i), limit ₂ (i) and S ^F (m) The corresponding relationship between the time-domain samples in T _env (i) and the corresponding relationship between the time-domain samples in S ^F (m) is the same.

In the above post-processing process, a better threshold limit ₁ (i), limit ₂ (i) setting method is:

limit ₁ (i) = 2^T _env (i),

limit ₂ (i)=[2^T _env (i)]×2.5.

In addition, the above post-processing process can process every 80 sample points once, and divide each 80 sample points into three sections, the first 6 sample points (m ₁ ~ m ₂ part), the middle 70 sample points (m ₂ +1～m ₃ part), the last 4 sample points (m ₃ +1～m ₄ part). An example is as follows: if N=160, the signal after time-frequency adjustment is 320 samples, which can be divided into 4 times for post-processing; among them, the part of m ₁ ~m ₂ is 0~5, 80~85, 160~165 , 240~245 parts; m ₂ +1~m ₃ parts are 6~75, 86~155, 166~235, 246~315 parts; m ₃ +1~m ₄ parts are 76~79, 156 ～159, 236～239, 316～319 parts.

The following describes the reconstruction device of the medium and high frequency signal of the bandwidth extension of the embodiment of the present invention for performing the above high frequency signal reconstruction method. Referring to FIG. 4, its basic logical structure includes:

The excitation signal generation unit 201 adopts the logical structure of the excitation signal generation device of the foregoing embodiment, and is used to generate the excitation signal exc _HB (m), m=0,...,2N-1;

Decoding unit 202, used to decode and output time-domain spectral envelope parameters T _env (i) and frequency-domain spectral envelope parameters F _env (j), wherein i=0, ..., I-1, j=0, ..., J -1;

The time-domain shaping unit 203 is configured to adjust the time-domain spectral envelope of the exc _HB (m) output by the excitation signal generating unit 201 according to the T _env (i) output by the decoding unit 202, and adjust each T _env (i) correspondingly Exc _HB (m) includes a section of A time-domain samples, A≤2N/I, and outputs the time-domain adjusted signal S ^T (m);

The frequency domain shaping unit 204 is used to adjust the frequency domain spectrum envelope of ^ST (m) output by the time domain shaping unit 203 according to the F _env (j) output by the decoding unit 202, and adjust each F _env (j) correspondingly S ^T (m) is a sub-band with a bandwidth of B ₁ in the frequency domain, B ₁ ≤ _{B 2} /J, B ₂ is the frequency bandwidth of S ^T (m), output the reconstructed signal S ^F (m) after frequency domain adjustment ;

The spectrum folding unit 205 is configured to perform spectrum folding on the input S ^F (m) to generate a high-frequency reconstruction signal _SHB (m) with a frequency range of 2B ₀ -2B ₀ +B ₂ .

In addition, based on the post-processing process used to eliminate signal burrs in the aforementioned reconstruction method, the high-frequency signal reconstruction device of this embodiment may further include:

The post-processing unit 206 is configured to use the envelope adjustment thresholds limit ₁ (i), limit ₂ (i) to perform envelope adjustment on the ^SF (m) output by the frequency domain shaping unit 204, and the adjusted ^SF (m) is : In the part of m=m ₁ ~m ₂ , if |S ^F,old (m)|<limit ₁ (i), then S ^F (m)=S ^F,old (m); in m=m ₂ In the part of +1～m ₃ , if limit ₁ (i)≤|S ^{F, old} (m)|≤limit ₂ (i), then S ^F (m)＝[S ^{F, old} (m)-limit ₁ (i)]/2+limit ₁ (i); in the part of m=m ₃ +1～m ₄ , if |S ^{F, old} (m)|＞limit ₂ (i), then S ^F (m) ＝[S ^{F, old} (m)-limit ₂ (i)]/16+limit ₂ (i); where, S ^{F, old} (m) is S ^F (m) before envelope adjustment; limit ₁ (i ), the corresponding relationship between limit ₂ (i) and time-domain samples in S ^F (m) is the same as the corresponding relationship between T _env (i) and time-domain samples in S ^F (m); the adjusted S ^F (m) is output to the spectrum folding unit 205 .

The time-domain gain factor smooth interpolation method further provided in the above-mentioned high-frequency signal reconstruction method and device embodiment can obtain a better time-domain adjustment effect; the further provided specific frequency-domain spectral envelope adjustment method avoids the use of multiple The filter component frequency band filters the signal, which simplifies the processing process and reduces the computational complexity; the post-shaping processing method further provided can better eliminate the glitches in the shaping process.

In order to better understand the above-mentioned embodiment, an application in ultra-wideband bandwidth expansion is taken as an example below to illustrate the above-mentioned high-frequency signal reconstruction process:

① Generate 0-6kHz high-frequency excitation signal exc _HB (m), 320 samples per frame in the time domain. That is, 2N=320, B ₀ =4 kHz, B ₂ =3B ₀ /2=6 kHz.

② Obtain 32 time-domain spectral envelope parameters T _env (i) by decoding from the code stream, i=0,...,31, each corresponding to 10 time-domain samples, ie I=32, A=10.

③ Divide exc _HB (m) into 32 subsections, each with 10 samples, and calculate the corresponding T' _env (i):

${{\mathrm{<span\; class="notranslate">\; T</span>}}^{<span\; class="notranslate">\; \text{'}</span>}}_{\mathrm{<span\; class="notranslate">\; env</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \{</span>\underset{\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 9</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; exc</span>}}_{\mathrm{<span\; class="notranslate">\; HB</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; .</span>$

Then calculate the time-domain gain g _T (i) = 2^[T _env (i)-T' _env (i)], and interpolate each g _T (i) with a smooth interpolation algorithm:

g _T,i (a)=w _T (a)·g _T (i)+[1−w _T (a)]·g ^last _T,i (a),a=0, . . . ,4.

g _{T, i} (a) = g T (i), a = 5, . . . , 9.

Wherein, w _T (a)={0.0669872981f, 0.2500000000f, 0.5000000000f, 0.7500000000f, 0.9330127019f}, a is 0 to 4 in sequence, and f represents a floating point number. Then calculate the signal after time domain shaping:

S ^T (m) = g _{T, i} (a) · exc _HB (m).

④ Obtain 15 frequency-domain spectral envelope parameters F _env (j) by decoding from the code stream, j=0,...,14, each corresponding to a sub-band with a bandwidth of 0.4kHz, that is, J=15.

⑤ Add sine window w _TDAC (k) to S ^T (m) and the previous frame S ^{T, last} (m),

w _TDAC (k)=sin[(k+0.5)π/640], k=0, ..., 639; obtain the windowed signal S ^w (k),

S ^w (k) = w _TDAC (k) S ^{T, last} (k), k = 0, ..., 319,

S ^w (k) = w _TDAC (k) S ^T (k-2N), k = 320,..., 639;

Then perform 640-point MDCT on the windowed S ^w (k) sequence to generate a frequency domain signal S ^F1 (m),

${\mathrm{<span\; class="notranslate">\; S</span>}}^{\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 639</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; S</span>}}^{\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; cos</span>}<span\; class="notranslate">\; [</span>\frac{\mathrm{<span\; class="notranslate">\; \&pi;</span>}}{\mathrm{<span\; class="notranslate">\; 1280</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 320</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span><span\; class="notranslate">\; ;</span>$

Since 3/4 low-pass filtering is performed in the process of generating exc _HB (m), the data in the 6-8kHz frequency band is filtered out, so only the data in the 0-6kHz frequency band is valid, so the first 240 bits of S ^F1 (m) are extracted Points are used to calculate 15 F' _env (j), every 16 point group, namely D=16,

${{\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \text{'}</span>}}_{\mathrm{<span\; class="notranslate">\; env</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \{</span>\underset{\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; 15</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; S</span>}}^{\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; 16</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; .</span>$

Then calculate the frequency domain gain g _F (i)=2^[F _env (j)-F' _env (j)], and obtain the frequency domain shaped signal S ^F2 (m)=g _F (i)·S ^F1 ( m). Then perform IMDCT on S ^F1 (m) to obtain S ^F (m).

⑥The 320 samples of S ^F (m) are processed every 80 samples, and each time is divided into three sections, the first 6 samples, the middle 70 samples, and the last 4 samples, according to the limit ₁ (i )=2^T _env (i), limit ₂ (i)=[2^T _env (i)]×2.5 for envelope adjustment.

⑦Then perform spectrum folding on the signal of 0-6 kHz after envelope adjustment, and obtain the high-frequency reconstruction signal _SHB (m) of 8-14 kHz.

Combining the _SHB (m) with the low-frequency signal (0-8kHz) obtained by decoding the core code stream (for example, through QMF synthesis) can obtain a complete ultra-wideband reconstruction signal (0-14kHz).

Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above-mentioned embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium can include: ROM, RAM, disk or CD, etc.

The method for generating the excitation signal in the bandwidth extension provided by the present invention and the corresponding high-frequency signal reconstruction method and device have been introduced in detail above. In this paper, specific examples are used to illustrate the principle and implementation of the present invention. The above implementation The description of the example is only used to help understand the method of the present invention and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation and scope of application. In summary As stated above, the content of this specification should not be construed as limiting the present invention.

Claims (19)

1. the generation method of pumping signal is characterized in that during a bandwidth was expanded, and comprising:

The generated frequency scope is 0～B ₀The first pumping signal exc (n), n=0 ..., N-1;

Exc (n) is carried out spectrum folding, and the generated frequency scope is B ₀～2B ₀The second pumping signal exc ^Fold(n);

To exc (n) and exc ^Fold(n) carry out synthetic filtering, reference frequency output is 0～2B ₀The 3rd pumping signal exc _HB(m), m=0 ..., 2N-1, described the 3rd pumping signal exc _HB(m) be used for carrying out the reconstruction of high-frequency signal as high-frequency excitation signal.

2. the generation method of pumping signal according to claim 1 is characterized in that, and is described to exc (n) and exc ^Fold(n) step of carrying out synthetic filtering is specially: to exc (n) and exc ^Fold(n) carry out the orthogonal mirror image synthetic filtering.

3. the generation method of pumping signal according to claim 1 and 2 is characterized in that, also comprises:

To frequency range is 0～2B ₀Exc _HB(m) carry out 3/4 low-pass filtering, reference frequency output is 0～3B ₀/ 2 exc _HB(m).

4. the generation method of pumping signal according to claim 3 is characterized in that, the step of described generation exc (n) is specially:

The decoding core code stream obtains constant codebook excitations and adaptive codebook excitation and gain separately;

According to described constant codebook excitations of gain weighting superposition and adaptive codebook excitation acquisition exc (n) separately.

5. the generation method of pumping signal according to claim 4 is characterized in that:

Described constant codebook excitations comprises that basic layer constant codebook excitations c (n) and enhancement layer strengthen excitation c ' (n), and corresponding gain is respectively g _cAnd g _Enh

Calculate exc (n) according to following formula:

Exc (n)=g _pV (n)+g _cC (n)+g _EnhC ' (n) wherein, v (n) is adaptive codebook excitation, g _pBe the gain of v (n), N=160, B ₀=4kHz.

6. the method for reconstructing of a bandwidth expansion medium-high frequency signal is characterized in that, comprising:

Generate pumping signal exc according to any described method of claim 1～5 _HB(m), m=0 ..., 2N-1;

Decoding obtains time domain spectrum envelope parameter T _Env(i) and frequency domain spectra envelope parameters F _Env(j), i=0 wherein ..., I-1, j=0 ..., J-1;

According to T _Env(i) to exc _HB(m) time domain spectrum envelope is adjusted, each T _Env(i) the corresponding exc that adjusts _HB(m) comprise a section of A time domain sampling point in, A≤2N/I generates the adjusted signal S of time domain ^T(m);

According to F _Env(j) to S ^T(m) frequency domain spectra envelope is adjusted, each F _Env(j) the corresponding S that adjusts ^T(m) bandwidth is B in the frequency domain ₁A subband, B ₁≤ B ₂/ J, B ₂Be S ^T(m) frequency span generates the adjusted reconstruction signal S of frequency domain ^F(m);

To S ^F(m) carry out spectrum folding, the generated frequency scope is 2B ₀～2B ₀+ B ₂High-frequency reconstruction signal S _HB(m).

7. the method for reconstructing of high-frequency signal according to claim 6 is characterized in that, and is described according to T _Env(i) to exc _HB(m) step that time domain spectrum envelope is adjusted comprises:

Calculate T according to coding side _Env(i) mode is calculated exc _HB(m) time domain spectrum envelope parameter T ' _Env(i);

According to T _Env(i) and T ' _Env(i) the energy difference between is calculated the preliminary gain factor g of time domain _T(i), each g _T(i) corresponding to exc _HB(m) comprise a section of A time domain sampling point in;

Each g of interpolation _T(i) obtain A gain factor g _{T, i}(a), a=0 ..., A-1;

According to g _{T, i}(a) adjust exc _HBThe gain of the sampling point of A * I (m) obtains S ^T(m).

8. the method for reconstructing of high-frequency signal according to claim 7 is characterized in that, A=10, and I=N/5, described according to T _Env(i) and T ' _Env(i) the energy difference between is calculated g _T(i) step is specially:

g _T(i)＝2^[T _env(i)-T’ _env(i)]；

Each g of described interpolation _T(i) obtain A g _{T, i}(a) step is specially:

g _{T, i}(a)=w _T(a) g _T(i)+[1-w _T(a)] g ^Last _{T, i}(a); Wherein, w _T(a) be window function, work as a=0 ..., 4 o'clock, w _T(a)=1/2{1-cos[(a+1) π/6] }, work as a=5 ..., 9 o'clock, w _T(a)=1; g ^Last _{T, i}(a) be previous frame exc _HB(m) gain factor of corresponding sampling point.

9. according to the method for reconstructing of any described high-frequency signal of claim 6～8, it is characterized in that, described according to F _Env(j) to S ^T(m) step that frequency domain spectra envelope is adjusted comprises:

Calculate F according to coding side _Env(j) mode is to S ^T(m) carry out time-frequency conversion and generate frequency-region signal S ^F1(m) and calculate S ^F1(m) frequency domain spectra envelope parameters F ' _Env(j);

According to F _Env(j) and F ' _Env(j) the energy difference between is calculated the preliminary gain factor g of frequency domain _F(j), each g _F(j) corresponding to S ^F1(m) comprise a section of D frequency domain sampling point, D * J≤2N in;

Each g of interpolation _F(j) obtain D gain factor g _{F, j}(d), d=0 ..., D-1;

According to g _{F, j}(d) adjust S ^F1The gain of the sampling point of D * J (m) generates adjusted frequency-region signal S ^F2(m);

To S ^F2(m) carry out the inverse transformation of described time-frequency conversion, obtain S ^F(m).

10. the method for reconstructing of high-frequency signal according to claim 9 is characterized in that, and is described according to coding side calculating F _Env(j) mode generates S ^F1(m) and calculate F ' _Env(j) step comprises:

Be S ^T(m) and previous frame S ^{T, last}(m) windowing w _TDAC(k) the signal S after the acquisition windowing ^w(k), k=0 ..., 4N-1, wherein,

S ^w(k)＝w _TDAC(k)·S ^T，last(k)，k＝0，…，2N-1，

S ^w(k)＝w _TDAC(k)·S ^T(k-2N)，k＝2N，…，4N-1；

To S ^w(k) carry out discrete cosine transform and generate S ^F1(m),

Extract S ^F1(m) preceding D * J sampling point calculates F ' _Env(j),

11. the method for reconstructing of high-frequency signal according to claim 10 is characterized in that: D=16, J=3N/32, described window function w _TDAC(k) be:

w _TDAC(k)＝sin[(k+0.5)π/4N]。

12. the method for reconstructing according to any described high-frequency signal of claim 6～11 is characterized in that, to S ^F(m) carry out spectrum folding before, also comprise:

Use envelope to adjust threshold value limit ₁(i), limit ₂(i) to S ^F(m) carry out the envelope adjustment, adjusted S ^F(m) be:

At m=m ₁～m ₂Part in, if | S ^{F, old}(m) |＜limit ₁(i), S then ^F(m)=S ^{F, old}(m),

At m=m ₂+ 1～m ₃Part in, if limit ₁(i)≤| S ^{F, old}(m) |≤limit ₂(i), S then ^F(m)=[S ^{F, old}(m)-limit ₁(i)]/2+limit ₁(i),

At m=m ₃+ 1～m ₄Part in, if | S ^{F, old}(m) |＞limit ₂(i), S then ^F(m)=[S ^{F, old}(m)-limit ₂(i)]/16+limit ₂(i), wherein, S ^{F, old}(m) adjust preceding S for envelope ^F(m); Limit ₁(i), limit ₂(i) and S ^F(m) corresponding relation of time domain sampling point in, and T _Env(i) and S ^F(m) corresponding relation of time domain sampling point is identical in.

13. the method for reconstructing of high-frequency signal according to claim 12 is characterized in that: described limit ₁(i), limit ₂(i) be,

limit ₁(i)＝2^T _env(i)，limit ₂(i)＝[2^T _env(i)]×2.5。

14. the method for reconstructing of high-frequency signal according to claim 13 is characterized in that: N=160; Described m ₁～m ₂Part be 0～5,80～85,160～165,240～245 part; Described m ₂+ 1～m ₃Part be 6～75,86～155,166～235,246～315 part; Described m ₃+ 1～m ₄Part be 76～79,156～159,236～239,316～319 part.

15. the generating apparatus of pumping signal is characterized in that during a bandwidth was expanded, and comprising:

The core codec module, being used for reference frequency output is 0～B ₀The first pumping signal exc (n), n=0 ..., N-1;

The spectrum folding module is used for exc (n) is carried out spectrum folding, and reference frequency output is B ₀～2B ₀The second pumping signal exc ^Fold(n);

The synthetic filtering module is used for exc (n) and exc ^Fold(n) carry out synthetic filtering, reference frequency output is 0～2B ₀The 3rd pumping signal exc _HB(m), m=0 ..., 2N-1, described the 3rd pumping signal exc _HB(m) be used for carrying out the reconstruction of high-frequency signal as high-frequency excitation signal.

16. the generating apparatus of pumping signal according to claim 15 is characterized in that: described synthetic filtering module is the orthogonal mirror image composite filter.

17. the generating apparatus according to claim 15 or 16 described pumping signals is characterized in that, also comprises:

3/4 low-pass filter, being used for the incoming frequency scope is 0～2B ₀Exc _HB(m), it is carried out 3/4 low-pass filtering, reference frequency output is 0～3B ₀/ 2 exc _HB(m).

18. the reconstructing device of a bandwidth expansion medium-high frequency signal is characterized in that, comprising:

The pumping signal generation unit, the logical organization of the generating apparatus of any described pumping signal of employing claim 15～17 is used to generate pumping signal exc _HB(m), m=0 ..., 2N-1;

Decoding unit is used for decoding output time domain spectrum envelope parameter T _Env(i) and frequency domain spectra envelope parameters F _Env(j), i=0 wherein ..., I-1, j=0 ..., J-1;

The time domain shaping unit is used for according to T _Env(i) to exc _HB(m) time domain spectrum envelope is adjusted, each T _Env(i) the corresponding exc that adjusts _HB(m) comprise a section of A time domain sampling point in, A≤2N/I, the adjusted signal S of output time domain ^T(m);

The frequency-domain shaping unit is used for according to F _Env(j) to S ^T(m) frequency domain spectra envelope is adjusted, each F _Env(j) the corresponding S that adjusts ^T(m) bandwidth is B in the frequency domain ₁A subband, B ₁≤ B ₂/ J, B ₂Be S ^T(m) frequency span, the adjusted reconstruction signal S of output frequency domain ^F(m);

The spectrum folding unit is used for the S to input ^F(m) carry out spectrum folding, the generated frequency scope is 2B ₀～2B ₀+ B ₂High-frequency reconstruction signal S _HB(m).

19. the reconstructing device of high-frequency signal according to claim 18 is characterized in that, also comprises:

Post-processing unit is used to use envelope to adjust threshold value limit ₁(i), limit ₂(i) S that described frequency-domain shaping unit is exported ^F(m) carry out the envelope adjustment, adjusted S ^F(m) be: at m=m ₁～m ₂Part in, if | S ^{F, old}(m) |＜limit ₁(i), S then ^F(m)=S ^{F, old}(m); At m=m ₂+ 1～m ₃Part in, if limit ₁(i)≤| S ^{F, old}(m) |≤limit ₂(i), S then ^F(m)=[S ^{F, old}(m)-limit ₁(i)]/2+limit ₁(i); At m=m ₃+ 1～m ₄Part in, if | S ^{F, old}(m) |＞limit ₂(i), S then ^F(m)=[S ^{F, old}(m)-limit ₂(i)]/16+limit ₂(i); Wherein, S ^{F, old}(m) adjust preceding S for envelope ^F(m); Limit ₁(i), limit ₂(i) and S ^F(m) corresponding relation of time domain sampling point in, and T _Env(i) and S ^F(m) corresponding relation of time domain sampling point is identical in; With adjusted S ^F(m) export to described spectrum folding unit.

Images