HK40112969A - Time-domain superwideband bandwidth expansion for cross-talk scenarios - Google Patents

Description

Time-domain ultrawideband bandwidth extension for crosstalk scenarios

Technical Field

This disclosure relates to a method and apparatus for temporal bandwidth expansion of an excitation signal during encoding/decoding of a cross-talk audio signal.

In this disclosure and the appended claims:

The term "crosstalk" is generally intended to refer to a segment of speech in which a first sound element is superimposed on a second sound element, for example, but not exclusively, when the first person speaks on top of the second person.

The term "low frequency band" is intended to refer to a lower frequency range. Although frequency ranges of 0 kHz to 6.4 kHz and 0 kHz to 8 kHz are given as examples of "low frequency band" in this disclosure, the frequency boundaries of the low frequency band range can obviously be modified/adapted to the bit rate of the codec and/or to achieve specific objectives such as compliance with application, system, network, and design/business-related constraints.

The term "high frequency band" is intended to refer to a higher frequency range. Although frequency ranges of 6.4 kHz to 14 kHz and 8 kHz to 16 kHz are given as examples of "high frequency band" in this disclosure, the frequency boundaries of the high frequency band range can obviously be modified/adapted to the bit rate of the codec and/or to achieve specific objectives such as compliance with application, system, network, and design/business-related constraints.

Background Technology

In many conversational applications, it is common for one person to speak over another. As mentioned above, this situation is often referred to as “crosstalk.” Crosstalk can be problematic in modern speech coding/decoding systems. Since traditional speech coding techniques are primarily designed and optimized for single-speech content (only one person speaking), the quality of crosstalk can be severely affected by the coding/decoding operation. As an example, one of the most serious problems in crosstalk speech coding/decoding in the 3GPP EVS codec (reference [1] or the entire contents thereof are incorporated herein by reference) is the occasional presence of “click-click noise.” “Click-click noise” is a loud, annoying sound produced at frequencies from 8 kHz to 14 kHz (i.e., within the high-frequency band range example defined above in this document).

At the low bit rate of the 3GPP EVS codec, the high-band frequency content is encoded/decoded using the Ultra Wideband Bandwidth Extension (SWB TBE) tool as described in reference [1]. Due to the limited number of bits available to the SWB TBE tool, the high-band excitation signal in the high-band frequency range is not directly encoded. Instead, the low-band excitation signal in the low-band frequency range is computed using the ACELP (Algebraic Digital Excitation Linear Prediction) encoder (reference [2], the entire contents of which are incorporated herein by reference), and then upsampled and extended to 14 kHz or 16 kHz depending on the high-band frequency range, and used as a substitute for the high-band excitation signal. If there is a mismatch between the low-band excitation signal and the high-band excitation signal, the synthesized sound may sound different from the original sound. When the low-band excitation signal is voiced but the high-band excitation signal is unvoiced, the synthesized sound will be perceived as the click-clack noise defined above. The problem of click-clack noise in crosstalk content is illustrated in the spectrum diagram of Figure 1.

Figure 1 illustrates the power spectrum P versus frequency f of an exemplary crosstalk sound, where two speakers emit different types of sound. While the sound from the first speaker (speaker 1) primarily comprises audible content, the sound from the second speaker (speaker 2) contains silent segments. Assuming a mono capture device (such as a smartphone or omnidirectional microphone), the sounds from the two speakers will be mixed together within the capture device. As a result, as seen by the encoder, the spectral content of the input sound signal will resemble a superset of the two spectra. A similar situation occurs in multi-channel capture devices such as stereo microphones or surround microphones. If the encoder includes a downmixing module, the resulting mono input signal may contain different types of sound that are clearly distinguishable in the spectral domain.

Summary of the Invention

This disclosure relates to the following aspects:

- A method for time-domain bandwidth expansion of an excitation signal during decoding of a crosstalk audio signal, comprising: decoding a high-frequency band mixing factor received in a bitstream, and mixing a low-frequency band excitation signal and a random noise excitation signal using the high-frequency band mixing factor to generate a time-domain bandwidth expanded excitation signal.

- A method for temporal bandwidth expansion of an excitation signal during the encoding of a crosstalk audio signal, comprising: (a) calculating a high-frequency band residual signal using the audio signal and (b) calculating the time envelope of the high-frequency band residual signal; and calculating a high-frequency band voicing factor based on the time envelope of the high-frequency band residual signal.

- A method for time-domain bandwidth expansion of an excitation signal during the encoding of a crosstalk audio signal, comprising: calculating a high-frequency band mixing factor that can be used to mix a low-frequency band excitation signal and a random noise excitation signal to produce an excitation signal with time-domain bandwidth expansion.

- A method for temporal bandwidth expansion of an excitation signal during the encoding of a crosstalk audio signal, comprising: (a) calculating a high-frequency band residual signal using the audio signal and (b) calculating the time envelope of the high-frequency band residual signal; calculating a high-frequency band phonation factor based on the time envelope of the high-frequency band residual signal; calculating a high-frequency band mixing factor that can be used to mix a low-frequency band excitation signal and a random noise excitation signal to generate an excitation signal with temporal bandwidth expansion; and using the high-frequency band phonation factor to estimate a gain/shape parameter.

- An apparatus for time-domain bandwidth expansion of an excitation signal during decoding of a crosstalk audio signal, comprising: a decoder that decodes a high-frequency band mixing factor received in a bitstream, and a mixer that uses the high-frequency band mixing factor to mix a low-frequency band excitation signal and a random noise excitation signal to generate a time-domain bandwidth expanded excitation signal.

- An apparatus for time-domain bandwidth expansion of an excitation signal during the encoding of a crosstalk audio signal, comprising: (a) a calculator for calculating a high-frequency band residual signal using the audio signal and (b) a calculator for calculating the time envelope of the high-frequency band residual signal; and a calculator for a high-frequency band phonation factor based on the time envelope of the high-frequency band residual signal.

- An apparatus for time-domain bandwidth expansion of an excitation signal during the encoding of a crosstalk audio signal, comprising: a calculator for a high-frequency band mixing factor that can be used to mix a low-frequency band excitation signal and a random noise excitation signal to generate a time-domain bandwidth-expanded excitation signal.

- An apparatus for time-domain bandwidth expansion of an excitation signal during the encoding of a crosstalk audio signal, comprising: (a) a calculator for calculating a high-frequency band residual signal using the audio signal and (b) a calculator for calculating the time envelope of the high-frequency band residual signal; a calculator for a high-frequency band phonation factor based on the time envelope of the high-frequency band residual signal; a calculator for a high-frequency band mixing factor that can be used to mix a low-frequency band excitation signal and a random noise excitation signal to generate a time-domain bandwidth-expanded excitation signal; and an estimator for a gain/shape parameter using the high-frequency band phonation factor.

The foregoing and other objects, advantages and features of the method and apparatus for temporal bandwidth expansion of an excitation signal during the encoding/decoding of crosstalk audio signals will become more apparent after reading the following non-limiting description of illustrative embodiments given by way of example only with reference to the accompanying drawings.

Attached Figure Description

In the attached diagram:

Figure 1 is a graph showing the power spectrum P (dB) of an exemplary crosstalk sound emitting different types (audible and silent) of sound from two of the speakers (speaker 1 and speaker 2) versus frequency f (kHz).

Figure 2 is a schematic block diagram showing the calculation/calculator of the high-frequency band emission factor in a method and device for time-domain bandwidth expansion of the excitation signal during the encoding of crosstalk audio signals.

Figure 3 is a graph showing how to determine the time envelope of the high-frequency band residual signal;

Figure 4 is a graph showing the interpolation of the segment normalization factor calculated using the average value of the continuous segments of the downsampled time envelope of the high-frequency band residual signal.

Figure 5 is a schematic block diagram showing both the method for time-domain bandwidth expansion of the excitation signal at the decoder and the calculation/calculator for the time-domain bandwidth expansion of the excitation signal within the device.

Figure 6 is a schematic block diagram showing the calculation/calculator of the high-frequency band mixing factor formed/represented by the method for time-domain bandwidth expansion of the excitation signal and the quantized normalized gain in the device at the encoder.

Figure 7 is a schematic block diagram of a method and device for time-domain bandwidth expansion of excitation signals and a gain shape estimator/estimater.

Figure 8 is a graph showing the interpolation of subframe gain; and

Figure 9 is a simplified block diagram of an example configuration of hardware components for forming a method and apparatus for time-domain bandwidth extension of an excitation signal during encoding/decoding of a crosstalk audio signal.

Detailed Implementation

The following description relates to techniques for encoding/decoding crosstalk audio signals. In this disclosure, the encoding/decoding technique is based on the SWB TBE tool of the 3GPP EVS codec described in reference [1]. However, it should be noted that this technique can be used in combination with other encoding/decoding techniques.

More specifically, this disclosure proposes a series of modifications to the SWB TBE tool. The purpose of these modifications is to improve the quality of synthesized crosstalk audio signals (such as crosstalk speech signals), particularly, but not exclusively, to eliminate the click noise defined above. These modifications involve time-domain bandwidth expansion of the excitation signal and are distributed across one or more of the following three regions:

- In the encoder, the time envelope of the high-frequency band residual signal is used to calculate the high-frequency band phonation factor. In the SWBTBE tool, the high-frequency band corresponds to SHB (Ultra-High Frequency Band).

- Calculation of the high-frequency band mixing factor for the high-frequency band excitation signal in the encoder and decoder.

- Improvements in the estimation of gain/shape parameters and frame gain in the encoder and decoder.

The calculation of the high-frequency phonation factor according to this disclosure uses the high-frequency autocorrelation function itself, which is calculated, for example, from the time envelope of the high-frequency residual signal in the downsampling domain. The high-frequency phonation factor is used in the encoder instead of the so-called speech factor derived from the low-frequency phonation parameters in the SWB TBE tool.

The calculation of the high-frequency band mixing factor according to this disclosure replaces the corresponding method in the SWB TBE tool. The high-frequency band mixing factor determines the ratio of a low-frequency band excitation signal (e.g., from the ACELP core) to a random noise (which can also be defined as "white noise") excitation signal used to generate an excitation signal with temporal bandwidth expansion. In the disclosed implementation, for example in the downsampling domain, the high-frequency band mixing factor is calculated by minimizing the MSE (mean square error) between the temporal envelope of the random noise excitation signal and the temporal envelope of the low-frequency band excitation signal. Quantization of the high-frequency band mixing factor can be performed by an existing quantizer in the SWB TBE tool. Adding the quantized high-frequency band mixing factor to the SWB TBE bitstream causes a small increase in the bit rate. The mixing operation is performed at both the encoder and decoder. Other features of the mixing operation may include rescaling of the random noise excitation signal at the beginning of each frame and interpolation of the high-frequency band mixing factor to ensure a smooth transition between the current frame and the previous frame.

The estimation of the gain/shape parameters according to this disclosure includes: post-processing the gain/shape parameters (in the encoder) using adaptive smoothing of the unquantized gain/shape parameters by weighting the original gain/shape parameters with the interpolated gain/shape parameters. Quantization of the gain/shape parameters can be performed by the existing quantizer of the SWB TBE tool. Adaptive smoothing is applied twice; it is first applied to the unquantized gain/shape parameters (in the encoder), and then to the quantized gain/shape parameters (in both the encoder and decoder). Adaptive attenuation is applied to the unquantized frame gain at the encoder. The adaptive attenuation is based on the excess error (MSE), a byproduct of the SHB vocal parameter calculation in the SWB TBE tool.

Figure 2 is a schematic block diagram showing the calculation/calculator of the high-frequency band phonation factor in both the method 200 and the device 250 for time-domain bandwidth expansion of the excitation signal during the encoding of the crosstalk audio signal.

1. Low-frequency excitation signal

Referring to Figure 2, the input audio signal s _inp (n) to the 3GPP EVS codec can be represented, for example, by the following relation (1):

s _inp (n),n＝0,..,N _32k -1(1)

Where N _32k is the number of samples in a frame (frame length). In this particular non-limiting example, the input audio signal _sinp (n) is sampled at a rate of F _s = 32 kHz and the length of a single frame is N _32k = 640 samples. This corresponds to a time interval of 20 ms. Frames of a given duration, each comprising a given number of subframes and comprising a given number of consecutive audio signal samples, are used to process audio signals in the field of audio signal coding; further information about such frames can be found, for example, in reference [1].

Method 200 includes a downsampling operation 201, and device 250 includes a downsampler 251 for performing operation 201. The downsampler 251 downsamples the input audio signal _sinp (n) from 32 kHz to 12.8 kHz or 16 kHz depending on the encoder's bit rate. For example, for all bit rates up to 24.4 kbps, the input audio signal in the 3GPP EVS codec is downsampled to 12.8 kHz, otherwise it is downsampled to 16 kHz. The resulting signal is a low-frequency band signal 202. The low-frequency band signal 202 is encoded using the ACELP encoder 253 in ACELP encoding operation 203.

Method 200 includes ACELP encoding operation 203, while device 250 includes ACELP encoder 253 of 3GPP EVS codec to perform ACELP encoding. ACELP encoder 253 generates two types of excitation signals, adaptive codebook excitation signal 204 and fixed codebook excitation signal 205, as described in reference [1].

In method 200 and device 250, the SWB TBE tool within the 3GPP EVS codec performs a low-band excitation signal generation operation 207 and includes a corresponding generator 257 for generating a low-band excitation signal 208. Generator 257 takes two excitation signals 204 and 205 as inputs, mixes them together, and applies a nonlinear transformation to produce a mixed signal with an inverted spectrum, which is further processed in the SWB TBE tool to obtain the low-band excitation signal 208 of FIG2. Details regarding the generation of the low-band excitation signal can be found in reference [1]; specifically, Section 5.2.6.1 describes SWB TBE encoding, and Section 6.1.3.1 describes SWB TBE decoding.

As a non-limiting example, a low-frequency excitation signal 208 with a flipped spectrum is sampled at 16 kHz and expressed using the following relation (2):

l _LB (n), n＝0,..,N-1(2)

Where N = 320 is the frame length.

2. High-frequency band target signal

Referring to Figure 2, the high-frequency target signal 210 is essentially an extraction of the input audio signal _sinp (n), which contains spectral components in the frequency range of 6.4 kHz to 14 kHz or 8 kHz to 16 kHz, depending on the codec's bit rate. Regardless of the codec's bit rate, the high-frequency target signal 210 is always sampled at 16 kHz, and its spectral content is inverted. Therefore, the first frequency binary bit (bin) of the high-frequency target spectrum corresponds to the last frequency binary bit of the spectrum, and vice versa. In method 200 and device 250, the high-frequency target signal 210 can be generated, for example, using QMF (Quadrature Mirror Filter) analysis operation 209 performed by the QMF analysis filter bank 259 of the 3GPPEVS codec as described in reference [1]. Alternatively, the high-frequency target signal 210 can be generated by filtering the input audio signal _sinp (n) with a bandpass filter, shifting it in the frequency domain, inverting its spectral content as described above, and finally downsampling it from 32 kHz to 16 kHz. In this disclosure, it will be assumed that QMF processing is used, and the high-frequency target signal 210 will be represented, for example, by the following relation (3):

s _HB (n), n＝0,..,N-1(3)

Following processing in the QMF filter bank 259, method 200 includes an operation 211 for estimating the high-frequency band filter coefficients 212, and device 250 includes an estimator 261 for performing operation 211. The estimator 261 estimates the high-frequency band LP (linear prediction) filter coefficients 212 frame-by-frame from the high-frequency band target signal 210 in four consecutive subframes, each subframe having a length of 80 samples. The estimator 261 uses the Levinson-Durbin algorithm as described in reference [1] to compute the high-frequency band LP filter coefficients 212. The high-frequency band LP filter coefficients 212 can be represented using the following relation (4):

Where P = 10 is the order of the high-frequency band LP filter, and j = 0, ..., 3 are the subframe indices. The first LP filter coefficient in each subframe is unitary, i.e.

Method 200 includes an operation 213 of generating a high-frequency band residual signal 214, and device 250 includes a generator 263 for generating the high-frequency band residual signal to perform operation 213. Generator 263 generates the high-frequency band residual signal 214 by filtering the high-frequency band target signal 210 from QMF analysis filter bank 259 with a high-frequency band LP filter (LP filter coefficients 212) from estimator 261. The high-frequency band residual signal 214 can be represented, for example, using the following relation (5):

The first P samples of the high-frequency band residual signal 214 are calculated using the high-frequency band target signal 210 from the previous frame. This is indicated by the negative index in _sHB (-k), k = 1, ..., P in the summation term. The negative index refers to the sample of the high-frequency band target signal 214 at the end of the previous frame.

3. High-frequency autocorrelation function and phonation factor

Section 3 (High-Frequency Band Autocorrelation Function) deals with the characteristics of the encoder.

The high-frequency band residual signal 214, calculated by generator 263 using relation 5, is used to calculate the high-frequency band autocorrelation function and the high-frequency band phonation factor. The high-frequency band autocorrelation function is not directly calculated on the high-frequency band residual signal 214. Directly calculating the high-frequency band autocorrelation function requires significant computational resources. Furthermore, the dynamics of the high-frequency band residual signal 214 are typically low, and spectral reversal often leads to blurring the difference between audible and silent sound signals. To avoid these problems, the high-frequency band autocorrelation function is estimated, for example, on the time envelope of the high-frequency band residual signal 214 in the downsampling domain.

Method 200 includes an operation 215 of calculating the time envelope of the high-frequency band residual signal 214, and device 250 includes a calculator 265 for performing operation 215. To calculate the time envelope R _{_TD}(n) 216 of the high-frequency band residual signal 214, the calculator 265 processes the high-frequency band residual signal 214 through a sliding moving average (MA) filter, which in an example embodiment includes M = 20 taps. The time envelope calculation can be represented, for example, by the following relation (6):

Wherein, the negative sample r _HB (k), k = -M/2, ..., -1 refers to the value of the high-frequency band residual signal 214 in the previous frame. In the mode switching scenario, it is possible that the high-frequency band residual signal 214 in the previous frame has not been calculated and its value is unknown. In this case, the first M/2 value r _HB (k), k = 0, ..., M/2-1 is copied and used as a substitute for the value r _HB (k), k = -M/2, ..., -1 in the previous frame. The calculator 265 approximates the last M values of the time envelope R _TD (n) 216 in the current frame by IIR (Infinite Impulse Response) filtering. This can be done using the following relation (7):

R _TD (n)＝0.05·r _HB (n)+0.95·R _TD (n-1),n＝NM,...,N-1(7)

Figure 3 illustrates the operation 215 for calculating the time envelope R _TD (n) 216 of the high-frequency band residual signal 214.

Method 200 includes a time envelope downsampling operation 217, and device 250 includes a downsampler 267 for performing operation 217. Downsampler 267 downsamples the time envelope R_{_TD}(n)216 by a factor of 4 using, for example, the following relation (8):

R _4kHz (n)＝R _TD (4n),n＝0,...,N/4-1(8)

Method 200 includes an average calculation operation 219, and device 250 includes a calculator 269 for performing operation 219. Calculator 269 divides the downsampled time envelope R _4kHz (n) 218 into four consecutive segments and calculates the average value 220 of the downsampled time envelope R _4kHz (n) 218 in each segment using, for example, the following relation (9):

Where k is the index of the segment.

Calculator 269 limits all averages to the maximum value of 1.0.

Method 200 includes a normalization factor calculation operation 221, and device 250 includes a calculator 271 for performing operation 221. Calculator 271 uses a downsampled time envelope average 220 to calculate the segment normalization factor for each segment k using, for example, the following relation (10):

Then, calculator 271 uses, for example, the following relation (11), to linearly interpolate the segment normalization factor from relation (10) over the entire interval of the current frame to produce an interpolated normalization factor 222:

The interpolation process performed by operation 221 and calculator 271 is shown in Figure 4.

In relation (11), the term _η⁻¹ refers to the normalization factor of the last segment in the previous frame. Therefore, _η⁻¹ is updated with _η⁻³ after the interpolation process in each frame.

Method 200 includes a normalization operation 223 of the downsampled time envelope, and device 250 includes a normalizer 273 for performing operation 223. Normalizer 273 processes the downsampled time envelope R_{_4kHz}(n)₂₁₈ from downsampler 267 using, for example, the following relation (12): γ(n)₂₂₂, an interpolated normalization factor.

R _γ (n)＝R _4kHz (n)·γ(n),n＝0,...,N/4-1 (12)

Then, normalizer 273 subtracts the global average of the normalized envelope (relation (13)) from the value R _γ (n) of relation (12) to complete the normalization process of the downsampled temporal envelope in operation 223 (R _norm (n) 224 in Figure 2). This can be represented by relation (13):

It is useful to estimate the skewness of the time envelope of a high-frequency band residual signal. For this purpose, method 200 includes a time envelope skewness estimation operation 225, and device 250 includes an estimator 275 for performing operation 225. The time envelope skewness estimation can be accomplished by fitting a linear curve to the segment average calculated in relation (9) using the linear least squares (LLS) method. The skewness 226 of the time envelope is then the slope of the linear curve. The linear curve calculated using the LLS method is defined as:

According to the LLS method, the goal is to minimize the sum of squared differences between all k = 0, ..., 3. This can be expressed using the following relation (15):

The optimal slope a _LLS (slope 226) can be calculated by estimator 275 using relation (16):

Method 200 includes a high-band autocorrelation function calculation operation 227, and device 250 includes a calculator 277 for performing operation 227. Calculator 277 calculates the high-band autocorrelation function _Xcorr 228 based on a normalized time envelope, for example, relation (17):

Where E_{_f} is the energy of the normalized time envelope R_{_norm}(n)²²⁴ in the current frame, and is also the energy of the normalized time envelope R_{_norm}(n)²²⁴ in the previous frame. Calculator 277 can calculate the energy using the following relation (18):

In the case of mode switching, the factor preceding the summation term in relation (17) is set to 1/E _f , since the energy of the normalized time envelope R _norm (n)224 in the previous frame is unknown.

Method 200 includes a high-frequency band phonation factor calculation operation 229, and device 250 includes a calculator 279 for performing operation 229.

The emission of high-frequency band residual signals is closely related to the variance _σcorr of the high-frequency band autocorrelation function _Xcorr 228. Calculator 279 uses, for example, the following relationship (19) to calculate the variance _σcorr :

To improve the discrimination potential (sound/no sound determination) of the sound parameter v _mult , calculator 279 multiplies the variance σ _corr by the maximum value of the high-frequency autocorrelation function X _corr 228, as expressed by the following relationship (20):

Then, calculator 279 uses, for example, the following relation (21), to transform the vocal parameter _{v_mult} from relation (20) using a sigmoid function to limit its dynamic range, and obtains the high-frequency vocal factor _{v_HB} 230:

The factor β is estimated experimentally and set to, for example, a constant value of 25.0. Then, the high-frequency band phonation factor _vHB 230 calculated from the above relation (21) is limited to the range of <0.0;1.0> and sent to the decoder.

4. Excitation Mixture Factor

Figure 5 is a schematic block diagram showing the calculation/calculation of the excitation signal with time-domain bandwidth expansion in both method 200 and device 250 at the decoder.

Section 4 (Excitation Hybridity Factor) deals with the characteristics of both the encoder and decoder.

The SWB TBE tool in the 3GPP EVS codec uses the low-frequency band excitation signal 208 (Figure 2) described in Section 1 (Low-Frequency Excitation Signal) to predict the high-frequency band residual signal 214 (Figure 2) described in Section 2 (High-Frequency Band Target Signal). At lower bit rates (below 24.4 kbps) in the EVS codec, the SWB TBE tool uses 19 bits to encode the spectral envelope and energy of the predicted high-frequency band residual signal. For a frame length of 20 ms, this yields a bit rate of 0.95 kbps. At bit rates above 24.4 kbps, the SWB TBE tool uses 32 bits to encode the spectral envelope and energy of the predicted high-frequency band residual signal. For a frame length of 20 ms, this yields a bit rate of 1.6 kbps. At both bit rates (0.95kbps and 1.6kbps) of the SWB TBE tool, no bits are used to encode the high-frequency band residual signal 214 or the high-frequency band target signal 210.

Referring to Figure 5, method 200 includes a pseudo-random noise generation operation 501, and device 250 includes a pseudo-random noise generator 551 for performing operation 501.

The pseudo-random noise generator 551 generates a uniformly distributed random noise excitation signal 502. For example, the pseudo-random number generator of the 3GPP EVS codec described in reference [1] can be used as the pseudo-random noise generator 551. The random noise excitation signal w _rand 502 can be represented by the following relation (22):

w _rand (n)∈U[-32767;32768],n＝0,...,N-1 (22)

The random noise excitation signal w _rand 502 has zero mean and a non-zero variance σ _rand = 1.14e+11. It should be noted that the variance is only an approximation and represents the average value over 100 frames.

Method 200 includes operation 503 for calculating the power of the low-frequency band excitation signal l _LB (n) 208 and a power calculator 553 for performing operation 503.

The power calculator 503 uses, for example, the following relationship (23) to calculate the power 504 of the low-frequency band excitation signal l _LB (n) 208 sent from the encoder:

Method 200 includes an operation 505 that normalizes the power of a random noise excitation signal 502 and a power normalizer 555 for performing operation 505.

The power normalizer 555 normalizes the power of the random noise excitation signal 502 to the power 504 of the low-frequency band excitation signal 208, for example, using the following relationship (24):

Although the true variance of the random noise excitation signal 502 varies frame by frame, power normalization does not require an exact value. Instead, an approximation of the variance defined above is used in the above relation (24) to save computational resources.

Method 200 includes an operation 507 of mixing a low-frequency band excitation signal l _LB (n) 208 with a power-normalized random noise excitation signal w _white (n) 506, and a mixer 557 for performing operation 507.

Mixer 557 generates a time-domain bandwidth-extended excitation signal 508 by mixing the low-frequency band excitation signal l _LB (n) 208 with the power-normalized random noise excitation signal w _white (n) 506 using a high-frequency band mixing factor described later in this disclosure.

Figure 6 is a schematic block diagram showing the calculation/calculator of the high-frequency band mixing factor formed/represented by the method for time-domain bandwidth expansion of the excitation signal and the quantized normalized gain within the device at the encoder.

Referring to Figure 6, at the encoder,

Method 200 includes an operation 602 to calculate the time envelope of the power-normalized random noise excitation signal _wwhite (n)506, an operation 604 to calculate the time envelope of the low-frequency band excitation signal _lLB (n)208, an operation 601 to minimize mean square error (MSE), and an operation 607 to quantize gain.

and

- Device 250 includes a time envelope calculator 652 for performing operation 602, a time envelope calculator 654 for performing operation 604, an MSE minimizer 651 for performing operation 601, and a gain quantizer 657 for performing operation 607.

As shown in Figure 6, in order to save computational resources, the mean square error (MSE) minimization process is used to calculate the optimal gain based on the time envelope of the signal in the downsampled domain. Another advantage of this method is that it is more robust to background noise.

When calculating (operation 215 and calculator 265 in Figure 2) the time envelope of the high-frequency band residual signal 214 and downsampling that time envelope (operation 217 and downsampler 267 in Figure 2), calculator 652 uses the same algorithm described in Section 3 (High-Frequency Band Autocorrelation Function and Speech Factor) to calculate the downsampled time envelope _w4kHz (n)606 of the power-normalized random noise excitation signal _wwhite (n)506 (which is also calculated at the encoder shown in Figure 5 and the corresponding description). The downsampling factor used is, for example, 4. The downsampled time envelope of the power-normalized random noise excitation signal can be expressed by the following relation (25):

W _4kHz (n),n＝0,...,N/4-1(25)

Similarly, calculator 654 again uses the same algorithm described in Section 3 (High-Frequency Band Autocorrelation Function and Speech Factor) to calculate the time envelope L4kHz(n)605 of the low-frequency band excitation signal _lLB (n)208 downsampled to _4kHz . The downsampled time envelope 606 of the low-frequency band excitation signal _lLB (n)208 can be expressed as follows:

L _4kHz (n),n＝0,...,N/4-1(26)

The purpose of the MSE minimization operation 601 is to find the optimal gain that minimizes the energy of the error between (a) the combined time envelope (L _4kHz (n), W _4kHz (n)) and (b) the time envelope R _4kHz (n) of the high-frequency band residual signal r _HB (n)214. This can be mathematically expressed using the relation (27):

To this end, the MSE minimizer 651 solves the system of linear equations. Solutions can be found in scientific literature. For example, the optimal gain pair can be calculated using relation (28).

The values of _c0 , ..., _c4 and _c5 are given by the following formula.

Then, the MSE minimizer 651 uses, for example, the following relation (30) to calculate the minimum MSE error energy (excess error):

For further processing, the gain quantizer 657 scales the optimal gain in such a way that the gain g _ln associated with the time envelope L _4kHz (n) 605 of the low-frequency band excitation signal l _LB (n) becomes a unit 1 of the gain g _wn associated with the time envelope W _4kHz (n) 606 of the power-normalized random noise excitation signal w _white (n) 506, given using, for example, the following relation (31):

The result/advantage of rescaling relation (31) is that only one parameter (normalized gain g _{_wn} ) needs to be encoded and decoded and sent from the encoder to the decoder in the bitstream, instead of two parameters. Therefore, scaling the gain using relation (31) reduces bit consumption and simplifies the quantization process. On the other hand, the energy of the combined time envelope (L _{_4kHz} (n) and W _{_4kHz} (n)) will not match the energy of the time envelope R _{_4kHz}(n) of the high-frequency band residual signal 2¹⁴. This is not a problem because the SWBTBE tool uses the global gain and subframe gain, which contain information about the energy of the high-frequency band residual signal. The calculation of the subframe gain and global gain is described in Section 6 (Gain/Shape Estimation) of this disclosure.

Gain quantizer 657 limits the normalized gain _gwn to between a maximum threshold of 1.0 and a minimum threshold of 0.0. Gain quantizer 657 uses a 3-bit uniform scalar quantizer, for example, described by the following relation (32), to quantize the normalized gain _gwn :

The resulting index idx _g 610 is restricted to forming/representing the interval <0, 7> of the high-frequency band mixing factor, and is transmitted in the SWB TBE bitstream at 0.95kbps or 1.6kbps along with the existing index of the SWB TBE encoder.

Referring back to Figure 5, method 200 includes a hybrid factor decoding operation 509 at the decoder, and device 250 includes a hybrid factor decoder 559 for performing operation 509.

The hybrid factor decoder 559 uses, for example, the following relation (33) to generate the decoded gain from the received index idx _g 610:

The decoded gain from relation (33) forms the high-frequency band mixing factor _{f_mix} 510.

For example, a low-frequency band excitation signal l _LB (n) 208 sampled at 16 kHz and a normalized random noise excitation signal w _white (n) 506 sampled at 16 kHz are mixed together in mixer 557. However, the energy of both the low-frequency band excitation signal l _LB (n) 208 and the random noise excitation signal w _rand 502 varies frame by frame. If the low-frequency band excitation signal l _LB (n) 208 and the random noise excitation signal w _rand 502 are directly mixed using the high-frequency band mixing factor f _mix 510 obtained from relation (33), the energy fluctuations may eventually produce audible artifacts at frame boundaries. To ensure a smooth transition, the energy of the random noise excitation signal w _rand 502 is linearly interpolated between the previous frame and the current frame in generator 551. This can be done by scaling the random noise excitation signal w _rand 502 in the first half of the current frame with the following interpolation factor:

Where E _LB is the energy of the low-frequency band excitation signal l _LB (n)208 in the current frame and the energy of the low-frequency band excitation signal l _LB (n)208 in the previous frame.

To further smooth the transition between the previous and current frames, decoder 559 also performs linear interpolation on the high-frequency band mixing factor _{f_mix} 510. This can be done by introducing a scaling factor _{β_mix} (n) calculated, for example, using the following relation:

Where is the value of the high-frequency band mixing factor in the previous frame. Note that the interpolation factor _ζw (n) calculated in relation (34) and the scaling factor _βmix (n) calculated in relation (35) are defined for n = 0, ..., N/2-1.

The mixing of the low-frequency band excitation signal l _LB (n) 208 and the random noise excitation signal w _white (n) 506 is finally accomplished by the mixer 557 using, for example, relation (36) to obtain the time-domain bandwidth-extended excitation signal u (n) 508.

5. High-frequency bandgap synthesis (LP synthesis)

In the encoder of the SWB TBE tool, the high-frequency band LP filter coefficients 212, calculated by LP analysis of the high-frequency band input signal _sHB (n) in relation (4), are converted into LSF parameters and quantized. At a bit rate of 0.95 kbps, the SWB TBE encoder uses 8 bits to quantize the LSF index. At a bit rate of 1.6 kbps, the SWB TBE encoder uses 21 bits to quantize the LSF index.

Referring back to Figure 5, at the encoder,

Method 200 includes a decoding operation 511, and device 250 includes a corresponding decoder 561 for decoding the quantized LSF index; and

Method 200 includes a conversion operation 513, and device 250 includes a corresponding converter 563 for converting the decoded LSF index 512 into high-frequency band LP filter coefficients 514.

The decoded high-frequency band LP filter coefficients 512 can be expressed as:

Where P = 10 is the order of the LP filter. The first decoded LP filter coefficient in each subframe is a unit 1, i.e.

Method 200 includes a filtering operation 515, and device 250 includes a corresponding synthesis filter 565 that uses decoded high-frequency band LP filter coefficients 514 to filter the mixed time-domain bandwidth-extended excitation signal 508 of relation (36) using, for example, the following relation (38), to obtain the LP-filtered high-frequency band signal _yHB 516:

6. Gain/Shape Estimation (Figure 7)

Apply gain/shape parameter smoothing at both the encoder and decoder. Apply adaptive attenuation of frame gain only at the encoder.

The spectral shape of the high-frequency band target signal _sHB (n)210 is encoded using quantized LSF coefficients. Referring to Figure 7, the SWB TBE tool also includes an estimation operation 701/estimater 751 for estimating the time subframe gain 702 of the high-frequency band target signal _sHB (n)210 as described in reference [1]. Estimator 751 normalizes the estimated time subframe gain to unit energy.

The normalized estimated temporal subframe gain 702 from estimator 751 can be expressed using relation (39):

g _k , k＝0,...,3 (39)

Method 200 includes a computation operation 703, and device 250 includes a corresponding calculator 753 for determining the temporal slope 704 of the normalized estimated temporal subframe gain _gk 702 by linear least squares (LLS) interpolation. As shown in FIG8, this interpolation process can be accomplished by fitting a linear curve 801 to the true subframe gain 702 in four consecutive subframes (subframes 0 to 3 in FIG8) and calculating its slope.

The linear curve 801 constructed using the LLS interpolation method can be defined using the following relation (40):

The parameters c _{_LLS} and d_{_LLS} are found by minimizing the sum of squared differences between the actual subframe gain g _{_k}702 and the corresponding points on the linear curves of all k = 0, ..., 3 subframes. This can be expressed by the following relation (41):

By expanding relation (41), the time tilt _{g_tilt} of the estimated temporal subframe gain _{g_k} 702 can be expressed. The time tilt _{g_tilt} 702 is actually equal to the optimal slope _{c_LLS} of the linear curve. The time tilt _{g_tilt} can be calculated in calculator 753 using the following relation (42):

Method 200 includes a smoothing operation 705, and device 250 includes a corresponding smoother 755 for smoothing the temporal subframe gain _gk using the interpolated (LLS) gain from relation (40) when, for example, the following condition is true:

v _HB ＜0.4ANDidx _g ≥5AND|g _tilt |＜0.2 (43)

The smoothing of the temporal subframe gain _gk 702 is then accomplished by the smoother 755 using, for example, the following relation (44):

The weight κ is proportional to the vocal parameter v _HB 230 (Figure 2) given by relation (21). For example, the weight κ can be calculated using the following relation (45):

And it is limited to a maximum value of 1.0 and a minimum value of 0.0.

Method 200 includes a gain shape quantization operation 707, and device 250 includes a corresponding gain shape quantizer 757 for quantizing the smoothed temporal subframe gain weight 706. For this purpose, a gain shape quantizer of an encoder using, for example, a 5-bit SWB TBE tool as described in reference [1] can be used as quantizer 757. The quantized temporal subframe gain 708 from quantizer 757 can be expressed using the following relation (46):

Method 200 includes an interpolation operation 709, and device 250 includes a corresponding interpolator 759 for interpolating the quantized temporal subframe gain 708 again after the quantization operation 707 using the same LLS interpolation procedure as described in relations (40) and (41). The interpolated quantized subframe gain 710 in four consecutive subframes of a frame can be represented by the following relation (47):

Method 200 includes a tilt calculation operation 711, and device 250 includes a corresponding tilt calculator 761 for calculating the tilt of the interpolated quantized temporal subframe gain 710 using, for example, relation (42). The tilt of the interpolated quantized temporal subframe gain 710 can be expressed as:

Then, the quantized temporal subframe gain 708 is smoothed when the following condition (48) is true, where idx _g is an index from relation (32):

To this end, method 200 includes a quantization gain smoothing operation 713, and device 250 includes a corresponding smoother 714 for smoothing the quantization time subframe gain 708 by averaging, for example, the interpolated time subframe gain 710 from relation (47). For this purpose, the following relation (49) can be used:

Method 200 includes a frame gain estimation operation 715, and device 250 includes a corresponding frame gain estimator 765. The SWBTBE tool uses frame gain to control the global energy of the synthesized high-frequency band audio signal. The frame gain is estimated by multiplying the LP-filtered high-frequency band signal _yHB 516 of relation (a) by the energy match between the smoothed quantized time subframe gain 714 from relation (49) and the high-frequency band target signal _sHB (n) 210 of relation (b) using relation (3). The LP-filtered high-frequency band signal _yHB 516 of relation (38) is multiplied by the smoothed quantized time subframe gain 714 using, for example, the following relation (50):

The details of the frame gain estimation operation 715 are described in reference [1]. The estimated frame gain parameter is denoted as g _f (see 716).

Method 200 includes an operation 717 for calculating the synthesized high-frequency band signal 718, and device 250 includes a calculator 767 for performing operation 717. The calculator 767 can modify the estimated frame gain _gf 717 under certain conditions. For example, given the high-frequency band phonation factor _vHB 230 (Figure 2) and the MSE excess error energy E _err , the frame gain _gf can be attenuated according to relation (51), as shown in relation (51):

g _f ←f _att ·g _f ,ifv _HB ＞0.1ANDE _err ＞5.0(51)

Where E _{_err} is the excess error energy of MSE calculated in relation (30), and f _{_att} is the attenuation factor, for example, calculated as:

f _att =1.0-0.04(E _err -5.0)(52)

Reference [1] describes further modifications to the frame gain _gf under certain specific conditions.

Then, the calculator 767 uses the frame gain quantizer of the encoder of the SWB TBE tool in reference [1] to quantize the modified frame gain.

Finally, calculator 767 uses, for example, the following relationship (53) to determine the synthesized high-frequency band sound signal 718:

7. Example configuration of hardware components

Figure 9 is a simplified block diagram of an example configuration of the hardware components of the aforementioned method 200 and device 250 (hereinafter referred to as "method 200 and device 250") for performing time-domain bandwidth extension of the excitation signal during encoding/decoding of the crosstalk signal.

Method 200 and device 250 may be implemented as part of a mobile terminal, a portable media player, or in any similar device. Device 250 (identified as 900 in FIG9) includes input 902, output 904, processor 906, and memory 908.

Input 902 is configured to receive an input signal. Output 904 is configured to provide an excitation signal with time-domain bandwidth extension. Input 902 and output 904 can be implemented in a common module (e.g., a serial input/output device).

Processor 906 is operatively connected to input 902, output 904, and memory 908. Processor 906 is implemented as one or more processors for executing code instructions to support the functions of various operations and elements of the methods 200 and apparatus 250 described above, as shown in the accompanying drawings and/or as described in this disclosure.

Memory 908 may include non-transitory memory for storing code instructions executable by processor 906. Specifically, processor-readable memory includes/stores non-transitory instructions that, when executed, cause the processor to implement the operations and elements of method 200 and device 250. Memory 908 may also include random access memory or buffers for storing intermediate processing data from various functions performed by processor 908.

Those skilled in the art will recognize that the description of method 200 and apparatus 250 is merely illustrative and not intended to be limiting in any way. Other embodiments will readily conceive of those skilled in the art upon receiving this disclosure. Furthermore, the disclosed method 200 and apparatus 250 can be customized to provide valuable solutions to existing needs and problems in encoding and decoding sound.

For clarity, not all conventional features of the implementation of method 200 and device 250 are shown or described. It should be understood, of course, that in the development of any such actual implementation of method 200 and device 250, many implementation-specific decisions may need to be made to achieve specific developer objectives, such as compliance with application, system, network, and business-related constraints, and these specific objectives will vary from one implementation to another and from one developer to another. Furthermore, it should be understood that the development work may be complex and time-consuming, but remains a routine engineering task for those skilled in the art of sound processing who will benefit from this disclosure.

According to this disclosure, the elements, processing operations, and/or data structures described herein can be implemented using various types of operating systems, computing platforms, network devices, computer programs, and/or general-purpose machines. Furthermore, those skilled in the art will recognize that less general-purpose devices, such as hardwired devices, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), etc., can also be used. Where a method comprising a series of operations and sub-operations is implemented by a processor, computer, or machine, and those operations and sub-operations can be stored as a series of non-transitory code instructions readable by the processor, computer, or machine, such non-transitory code instructions can be stored on a tangible and/or non-transitory medium.

The processing operations and elements of the method 200 and device 250 described herein may include software, firmware, hardware, or any combination of software, firmware, or hardware suitable for the purposes described herein.

In method 200 and device 250, various processing operations and sub-operations can be performed in various orders, and some of the processing operations and sub-operations can be optional.

Although the present disclosure has been described above by way of its non-limiting illustrative embodiments, these embodiments may be modified in any way within the scope of the appended claims without departing from the spirit and essence of the present disclosure.

8. References

This disclosure references the following sources, the entire contents of which are incorporated herein by reference:

[1] 3GPP TS26.445, “EVS Codec Detailed Algorithmic Description”, 3GPP Technical Specification (Version 12) (2014) - Sections 5.2.6.1 and 6.1.3.1.

[2] Bessette, B., Lefebvre, R., Salami, R. et al., “Techniques for high-quality ACELP coding of wideband speech”. Scandinavian International Conference EUROSPEECH 2001, September 3-7, 2001, Aalborg, Denmark, the 7th European Conference on Speech Communication and Technology, the 2nd INTERSPEECH event.

Claims (72)

1. A method for time-domain bandwidth expansion of an excitation signal during decoding of a crosstalk audio signal, comprising: Decoding the high-frequency band mixing factor received in the bitstream; and The high-frequency band mixing factor is used to mix the low-frequency band excitation signal and the random noise excitation signal to generate a time-domain bandwidth-extended excitation signal. 2. The method of claim 1, wherein decoding the high-frequency band mixing factor comprises decoding a quantized normalized gain received in the bitstream, and using the decoded quantized normalized gain to calculate the high-frequency band mixing factor. 3. The method according to claim 1 or 2, further comprising: interpolating the energy of the random noise excitation signal between a previous frame and a current frame of the audio signal to smooth the transition between the previous frame and the current frame. 4. The method of claim 3, further comprising: scaling a portion of the random noise signal in order to interpolate the energy of the random noise excitation signal. 5. The method according to any one of claims 1 to 4, comprising: interpolating the high-frequency band mixing factor between a previous frame and a current frame of the audio signal to ensure a smooth transition between the previous frame and the current frame. 6. The method according to any one of claims 1 to 4, comprising: estimating the quantized gain/shape parameters. 7. A method for temporal bandwidth expansion of an excitation signal during the encoding of a crosstalk audio signal, comprising: (a) Calculate the high-frequency band residual signal using the sound signal and (b) Calculate the time envelope of the high-frequency band residual signal; The high-frequency band sound generation factor is calculated based on the time envelope of the high-frequency band residual signal. Calculate the high-frequency mixing factor that can be used to mix low-frequency band excitation signals and random noise excitation signals to generate excitation signals with extended time-domain bandwidth; and The high-frequency band phonation factor is used to estimate the gain/shape parameters. 8. A method for time-domain bandwidth expansion of an excitation signal during the encoding of a crosstalk audio signal, comprising: (a) Calculate the high-frequency band residual signal using the sound signal and (b) Calculate the time envelope of the high-frequency band residual signal; and The high-frequency band sound generation factor is calculated based on the time envelope of the high-frequency band residual signal. 9. A method for time-domain bandwidth expansion of an excitation signal during the encoding of a crosstalk audio signal, comprising: Calculate the high-frequency mixing factor that can be used to mix low-frequency band excitation signals and random noise excitation signals to generate excitation signals with extended time-domain bandwidth. 10. The method of claim 7 or 8, wherein calculating the high-frequency band phonation factor comprises (a) calculating a high-frequency band autocorrelation function based on the time envelope, and (b) using the high-frequency band autocorrelation function to calculate the high-frequency band phonation factor. 11. The method according to any one of claims 7, 8 and 10, wherein calculating the high-frequency band phonation factor comprises downsampling the time envelope of the high-frequency band residual signal by a given factor. 12. The method of claim 11, wherein calculating the high-frequency band phonation factor comprises dividing the downsampled time envelope into multiple segments and calculating the average value of each segment of the downsampled time envelope. 13. The method of claim 12, wherein calculating the high-frequency band phonation factor includes normalizing each segment of the downsampled time envelope of the high-frequency band residual signal. 14. The method of claim 13, wherein the normalization of each segment of the downsampled temporal envelope comprises (a) calculating a segment normalization factor based on a calculated average value, (b) interpolating the segment normalization factor in the current frame, and (c) normalizing the downsampled temporal envelope using the interpolated segment normalization factor. 15. The method according to any one of claims 7, 8 and 10 to 14, comprising: calculating the tilt of the time envelope of the high-frequency band residual signal based on linear least squares. 16. The method of claim 14, wherein calculating the high-frequency band phonation factor comprises (a) calculating a high-frequency band autocorrelation function based on a normalized time envelope, and (b) using the high-frequency band autocorrelation function to calculate the high-frequency band phonation factor. 17. The method according to any one of claims 7 and 9 to 16, wherein calculating the high-frequency band mixing factor comprises calculating and quantizing the gain from which the high-frequency band mixing factor is obtained. 18. The method of claim 17, wherein calculating the high-frequency band mixing factor includes generating a random noise excitation signal. 19. The method of claim 18, wherein generating the random noise excitation signal includes power normalization of the random noise excitation signal. 20. The method of claim 18 or 19, wherein calculating the high-frequency band mixing factor comprises (a) mixing the low-frequency band excitation signal with the random noise excitation signal, and (b) minimizing the mean square error between the mixed excitation signal and the high-frequency band residual signal calculated based on the sound signal. 21. The method according to any one of claims 18 to 20, wherein calculating the high-frequency band mixing factor comprises (a) calculating the time envelope of the random noise excitation signal, (b) calculating the time envelope of the low-frequency band excitation signal, and (c) finding the corresponding gains of the time envelopes of the random noise excitation signal and the low-frequency band excitation signal by means of a mean square error minimization process. 22. The method of claim 21, wherein calculating the high-frequency band mixing factor comprises scaling the gain of the time envelope of the random noise excitation signal and the low-frequency band excitation signal. 23. The method of claim 22, wherein scaling the gain includes obtaining a single gain parameter, and wherein calculating the high-frequency band mixing factor includes quantizing the single gain parameter to obtain the quantized gain, and obtaining the high-frequency band mixing factor from the quantized gain. 24. The method of claim 7 or 8, further comprising: using the high-frequency band phonation factor to estimate the gain/shape parameter. 25. The method according to any one of claims 7 to 24, wherein the gain/shape parameter is selected from the group consisting of: - The spectral shape of the high-frequency target signal; - Subframe gain of high-frequency target signals; - Frame gain parameter. 26. The method according to any one of claims 7 to 25, wherein estimating the gain/shape parameter includes calculating the time skewness of the gain/shape parameter. 27. The method of claim 26, wherein calculating the time tilt includes interpolating the gain/shape parameter. 28. The method of claim 27, wherein interpolating the gain/shape parameter comprises using linear least squares. 29. The method of any one of claims 7 to 28, wherein estimating the gain/shape parameter includes smoothing the gain/shape parameter using an adaptive weighting parameter. 30. The method of claim 29, further comprising: using the high-frequency band phonation factor to calculate the adaptive weighting parameter. 31. The method of claim 29 or 30, further comprising: smoothing the gain/shape parameters using the adaptive weighting parameters in response to a given condition relating to the high-frequency band phonation factor. 32. The method according to any one of claims 29 to 31, wherein estimating the gain/shape parameter includes quantizing the smoothed gain/shape parameter. 33. The method of claim 32, wherein estimating the gain/shape parameter includes interpolating the quantized gain/shape parameter. 34. The method of claim 32 or 33, wherein estimating the gain/shape parameter includes smoothing the quantized gain/shape parameter. 35. The method of claim 34, wherein smoothing of the quantized gain/shape parameters is performed by averaging the quantized interpolated gain/shape parameters. 36. The method of any one of claims 7 to 35, wherein estimating the gain/shape parameter includes adaptive attenuation of the frame gain parameter using MSE excess error. 37. An apparatus for performing time-domain bandwidth expansion of an excitation signal during decoding of a crosstalk audio signal, comprising: A decoder that decodes the high-frequency band mixing factors received in a bitstream; and A mixer that uses the high-frequency mixing factor to mix low-frequency excitation signals and random noise excitation signals to generate a time-domain bandwidth-extended excitation signal. 38. The apparatus of claim 37, wherein the decoder of the high-frequency band mixing factor decodes the quantized normalized gain received in the bitstream and uses the decoded quantized normalized gain to calculate the high-frequency band mixing factor. 39. The apparatus of claim 37 or 38, further comprising: a generator of the random noise excitation signal, which interpolates the energy of the random noise excitation signal between a previous frame and a current frame of the audio signal to smooth the transition between the previous frame and the current frame. 40. The method of claim 39, further comprising: in order to interpolate the energy of the random noise excitation signal, the generator of the random noise excitation signal scales the random noise signal in a portion of the current frame. 41. The device according to any one of claims 37 to 40, wherein the decoder of the high-frequency band mixing factor interpolates the high-frequency band mixing factor between a previous frame and a current frame of the audio signal to ensure a smooth transition between the previous frame and the current frame. 42. The device according to any one of claims 37 to 40, comprising: an estimator for estimating quantized gain/shape parameters. 43. An apparatus for performing time-domain bandwidth expansion of an excitation signal during the encoding of a crosstalk audio signal, comprising: (a) A calculator for calculating the high-frequency band residual signal using the sound signal and (b) a calculator for calculating the time envelope of the high-frequency band residual signal; A calculator for calculating the high-frequency band sound generation factor based on the time envelope of the high-frequency band residual signal; A calculator for calculating the high-frequency mixing factor that can be used to mix low-frequency band excitation signals and random noise excitation signals to produce excitation signals with extended time-domain bandwidth; and An estimator for the gain/shape parameters is used to estimate the high-frequency band phonation factor. 44. An apparatus for performing time-domain bandwidth expansion of an excitation signal during the encoding of a crosstalk audio signal, comprising: (a) a calculator for calculating the high-frequency band residual signal using the sound signal and (b) a calculator for calculating the time envelope of the high-frequency band residual signal; and A calculator for calculating the high-frequency band sound generation factor based on the time envelope of the high-frequency band residual signal. 45. An apparatus for performing time-domain bandwidth expansion of an excitation signal during the encoding of a crosstalk audio signal, comprising: A calculator for calculating the high-frequency mixing factor that can be used to mix low-frequency band excitation signals and random noise excitation signals to produce excitation signals with extended time-domain bandwidth. 46. The device of claim 43 or 44, wherein the calculator for the high-frequency band emitting factor calculates the high-frequency band autocorrelation function based on the time envelope, and uses the high-frequency band autocorrelation function to calculate the high-frequency band emitting factor. 47. The device according to any one of claims 43, 44 and 46, wherein the calculator of the high-frequency band phonation factor comprises a downsampler that downsamples the time envelope of the high-frequency band residual signal by a given factor. 48. The device of claim 47, wherein the calculator for the high-frequency band phonation factor includes a divider for dividing the downsampled time envelope into multiple segments, and a calculator for calculating the average value of each segment of the downsampled time envelope. 49. The device of claim 48, wherein the calculator of the high-frequency band phonation factor includes a normalizer for each segment of the downsampled time envelope of the high-frequency band residual signal. 50. The apparatus of claim 49, wherein each segment normalizer (a) calculates a segment normalization factor based on the calculated average value, (b) interpolates the segment normalization factor in the current frame, and (c) uses the interpolated segment normalization factor to normalize the downsampled temporal envelope. 51. The device according to any one of claims 43, 44 and 46 to 50, comprising: a calculator for calculating the tilt of the time envelope of the high-frequency band residual signal based on linear least squares method. 52. The device of claim 50, wherein the calculator for the high-frequency band phonation factor comprises a calculator that calculates a high-frequency band autocorrelation function based on a normalized time envelope and uses the high-frequency band autocorrelation function to calculate the high-frequency band phonation factor. 53. The device according to any one of claims 43 and 45 to 52, wherein the calculator of the high-frequency band mixing factor calculates and quantizes the gain forming the high-frequency band mixing factor. 54. The device of claim 53, wherein the calculator with the high-frequency band mixing factor includes a generator of a random noise excitation signal. 55. The apparatus of claim 54, comprising: a power normalizer for the power of the random noise excitation signal to the low-frequency band excitation signal. 56. The device according to claim 54 or 55, wherein the calculator of the high-frequency band mixing factor (a) combines the low-frequency band excitation signal with the random noise excitation signal, and (b) minimizes the mean square error between the mixed excitation signal and the high-frequency band residual signal calculated based on the sound signal. 57. The device according to any one of claims 54 to 56, wherein the calculator of the high-frequency band mixing factor (a) includes a calculator of the time envelope of the random noise excitation signal and a calculator of the time envelope of the low-frequency band excitation signal, and (b) finds the corresponding gains of the time envelopes of the random noise excitation signal and the low-frequency band excitation signal by a mean square error minimization process. 58. The device of claim 57, wherein the calculator of the high-frequency band mixing factor scales the gain of the time envelope of the random noise excitation signal and the low-frequency band excitation signal. 59. The apparatus of claim 58, wherein, in order to scale the gain of the time envelope of the random noise excitation signal and the low-frequency band excitation signal, the calculator of the high-frequency band mixing factor calculates a single gain parameter and quantizes the single gain parameter to obtain a quantized gain forming the high-frequency band mixing factor. 60. The device of claim 43 or 44, comprising: an estimator for estimating gain/shape parameters using the high-frequency band phonation factor. 61. The device according to any one of claims 43 to 60, wherein the gain/shape parameter is selected from the group consisting of: - The spectral shape of the high-frequency target signal; - Subframe gain of high-frequency target signals; - Frame gain parameter. 62. The device according to any one of claims 43 to 61, wherein the gain/shape parameter includes the subframe gain of the high-frequency band target signal, and wherein the estimator of the gain/shape parameter includes a calculator for calculating the time tilt of the subframe gain. 63. The device of claim 62, wherein the calculator for the time tilt includes an interpolator that interpolates the subframe gain. 64. The device of claim 63, wherein the interpolator that interpolates the subframe gain uses linear least squares. 65. The apparatus according to any one of claims 43 to 64, wherein the gain/shape parameter includes the subframe gain of the high-frequency band target signal, and the estimator of the gain/shape parameter includes a smoother that uses adaptive weight parameters to smooth the subframe gain. 66. The device of claim 65, wherein the smoother of the subframe gain uses the high-frequency band phonation factor to calculate the adaptive weight parameters. 67. The device of claim 29 or 30, wherein the smoother of the gain/shape parameter performs smoothing of the gain/shape parameter using the adaptive weighting parameter in response to a given condition relating to the high-frequency band phonation factor. 68. The apparatus according to any one of claims 65 to 67, wherein the estimator of the gain/shape parameter comprises a quantizer that quantizes the subframe gain. 69. The apparatus of claim 68, wherein the estimator of the gain/shape parameter comprises an interpolator that interpolates the quantized subframe gain. 70. The device of claim 68 or 69, wherein the estimator of the gain/shape parameter includes a smoother for smoothing the subframe gain. 71. The apparatus of claim 70, wherein the smoother smoothing the subframe gain smooths the quantized gain/shape parameters by averaging the quantized interpolated gain/shape parameters. 72. The apparatus according to any one of claims 43 to 71, wherein the gain/shape parameter includes the subframe gain of the high-frequency band target signal, and wherein the estimator of the gain/shape parameter uses MSE excess error to perform adaptive attenuation of the frame gain parameter.