JPH043877B2 - - Google Patents

Description

[Detailed Description of the Invention] [Technical Field] The sound source information of an input audio signal is represented by a plurality of impulse sequences, so-called multi-pulses, and information about the spectral envelope of the input audio signal is transmitted from the encoding side (analysis side) to the decoding side. A multi-pulse encoding/decoding device that transmits an input audio signal to a synthesizer (synthesizing side) and reproduces the input audio signal has recently become well known.

This type of multi-pulse is capable of transmitting sound source information including waveform information with a much smaller number of impulses compared to sound source pulses, and basically transmits much more data bits than a normal pocoder. It has the characteristic of providing excellent reproduction sound quality.

As a means of forming multi-pulses, BS
A-b-S (Analysis) proposed by Atal et al.
-by-Synthesis) method and the forward method by Ozawa, Araseki, Ono, etc. are two representative methods.

The A-b-S method uses a sound source pulse train analyzed for one multipulse in each analysis frame.
Analysis and synthesis are repeated for each multi-pulse so that the synthesized speech signal synthesized using the LPC coefficients has a value closest to that of the input speech signal, and the multi-pulse train is divided into spectra for each analysis frame. This is determined based on area evaluation, and although the sound source pulse can be optimally modeled, there is a problem in that it requires complicated calculations for each multi-pulse.

On the other hand, forward processing is
This is a modification of the spectral domain evaluation by the S method to the correlation domain evaluation described below, and the degree of optimal modeling as a sound source pulse is not as good as the A-b-S method, which determines each multi-pulse one by one. The feature is that it is possible to form multi-pulses with a significant reduction in the amount of calculation processing required. The above-mentioned correlation domain evaluation is usually performed using the output S(Z)*W(Z) obtained by passing the input audio signal S(Z) through an auditory weighting filter with a transfer function W(Z), and the input audio signal using LPC (Linear Prediction). vocal tract filter coefficient H obtained by analyzing
(Z) is cross-correlated with the impulse response of the output H(Z)*W(Z) obtained through the auditory sound weighting filter W(Z), and a cross-correlation function is calculated for each analysis frame. Z)
*The residual waveform component obtained by taking the autocorrelation of the impulse response of W(Z) and subtracting the autocorrelation function from the cross-correlation function, that is, the one that minimizes the error component with the input audio signal waveform. This is an evaluation method in which the pulse is determined as a multi-pulse.

Here Z is exp(jλ) and λ=2πTΔ, ΔT
is the sampling period of the analysis frame, and is the frequency. Further, the symbol * represents a convolution integral.

Unlike the A-b-S method, multi-pulse determination based on such correlation region evaluation acyclically determines the multi-pulse to be determined for each analysis frame.
It can be determined in a so-called forward manner,
As mentioned above, this is an efficient method that can significantly reduce the amount of calculations.

However, in this type of conventional forward multipulse determination means, no restrictions are placed on the pulse intervals of adjacent multipulses, so the number of bits required to quantize the pulse intervals becomes large. It has the disadvantage of being stored away.

[Purpose of the invention]

An object of the present invention is to eliminate the above-mentioned drawbacks and provide a means for limiting the maximum interval between adjacent multipulses in a multipulse encoding/decoding device that determines multipulses in a forward manner. It is an object of the present invention to provide a multi-pulse encoding/decoding device that can significantly suppress consumption of the number of quantization bits in an interval.

[Structure of the invention]

The device of the present invention provides a forward () based on correlation area evaluation in which a multi-pulse representative of sound source information of an input audio signal is determined through cross-correlation between the input audio signal and the impulse response of a synthesis filter to be placed on the analysis side. A multi-pulse type encoding/decoding device that performs a search using forward processing is provided with multi-pulse interval limiting means for limiting the maximum interval between adjacent multi-pulses to a preset length.

〔Example〕

Next, the present invention will be explained in detail with reference to the drawings.

FIG. 1A is a block diagram showing the configuration of the encoding side in the first embodiment of the multi-pulse encoding/decoding device of the present invention, and FIG. 1B is a block diagram of the multi-pulse encoding/decoding device of the present invention. FIG. 3 is a block diagram showing the decoding configuration in the first embodiment.

The block diagram showing the embodiment on the encoding side shown in FIG. 6, an encoder 7 and a multiplexer 8, and the decoding side shown in FIG. 1B includes a demultiplexer 9,
decoders 10, 11, and LPC synthesizer 12,
It is configured to include a D/A (digital/analog) converter 13 and an LPF (Low Pass Filter) 14.

In FIG. 1A, an input audio signal S(Z) input via an input terminal 101 is supplied to a noise weighter 1 and an LPC analyzer 2. In FIG. The noise weighter 1 has a built-in noise weighting filter with a transfer function W(Z), and by passing the input audio signal through this filter, the convolution integral of S(Z)*W(Z) is executed, and the output is cross-correlated. It is supplied to the function calculator 3. Transfer function W of noise weighting filter
(Z) is supplied with the LPC coefficients obtained by LPC analysis of the input audio signal S(Z) by the LPC analyzer 2 which receives the input audio signal S(Z), the order thereof, and the perceptual weighting coefficients via the output line 201. It is set based on these data.

The LPC analyzer 2 quantizes the input audio signal S(Z) as a digital quantity with a preset number of bits for each analysis frame, and performs LPC analysis on this to obtain p-order K parameters (partial autocorrelation coefficients), etc. The LPC coefficients are supplied to the noise weighter 1 and are also supplied to the encoder 4 via an output line 202.

The encoder 4 quantizes and encodes the input LPC coefficients, and then sends them to the output line.
401 to the multiplexer 8 and decodes the quantized LPC coefficients to obtain the transfer function H.
The auditory weighted impulse response of the vocal tract filter of (Z), that is, the impulse response of H(Z)*W(Z) is obtained and sent to the cross-correlation function calculator 3 and autocorrelation function calculator 5 via the output line 402. supply

The cross-correlation calculator 3 uses H(Z) thus supplied.
The impulse response of *W(Z) and the auditory weighted input signal S(Z)*W(Z) are cross-correlated by convolution integration, and the calculated cross-correlation function is supplied to the multi-pulse searcher 6.

In addition, the autocorrelation function calculator 5 inputs H(Z)
*Calculate the autocorrelation function of the pulse response of W(Z) and supply it to the multi-pulse searcher 6.

Multi-pulse search usually uses the cross-correlation function and auto-correlation function supplied for each analysis frame to perform the calculation of equation (1) shown below to obtain a predetermined sound source pulse train and use it as a multi-pulse. However, in this embodiment, as will be described later, there is a limit on the maximum pulse interval for the multi-pulses obtained in this way.

In equation (1), m _i is i within the analysis frame.
The time position of the th pulse from the frame end, g _i is its amplitude, _hx (m _i ) is the cross-correlation function at time delay m _i , g _l is the amplitude of the l th pulse in the analysis frame, R _hh (|m _l −m _i |) is the autocorrelation function of the impulse response. Equation (1) shows that the amplitude g _i (m _i ) is optimal when emitting a multi-pulse at position m _i , and when searching for multi-pulses based on equation (1), Focusing on the position of the species, calculate the amplitude using equation (1), and find that the maximum absolute value of the amplitude is the sound source pulse that minimizes the residual waveform component, and by repeating the above process, The usual multipulse determining means is to find a plurality of sound source pulses and determine them as multipulses, but conventional multipulses determined in this way do not impose any restrictions on the interval between adjacent pulses. Not yet.

In this embodiment, a preset length limit is imposed on the interval between such multi-pulses by the following means.

In the first embodiment shown in FIGS. 1A and 1B, a multi-pulse is preliminarily searched, and the presence or absence of adjacent pulses with an interval that causes a limit is checked, and if a multi-pulse with an interval exceeding the limit exists, The pulse interval is equivalently set within the limit by any of the following means. This preliminary multi-pulse search is performed by a pulse interval check circuit built into the encoder 7, and if there is a pulse interval exceeding the limit, this information is sent to the multi-pulse searcher 6 via an output line 701. The pulse interval part has an amplitude of 0
(zero) pulse to represent this,
The amplitude 0 is given to the time slot of the corresponding part, one pulse with the minimum amplitude is removed by that amount, and the multi-pulse containing this is supplied to the encoder 7 to be encoded, and this is output to the output line.
702 to the multiplexer 8. If there is no multipulse interval exceeding the limit interval, the multipulse for each analysis frame is directly sent to the multiplexer 8 via the output line 702.
supplied to The above-mentioned time slots are a plurality of time regions obtained by subdividing the analysis frame in advance based on conditions such as sampling frequency and frame length.

Also, instead of setting a multi-pulse with an amplitude of 0 as described above, the maximum pulse in the vicinity of the time slot in which this multi-pulse should be set is searched and used for the same purpose as the multi-pulse with an amplitude of 0, and the minimum amplitude pulse can be set accordingly. It is also possible to supply the multipulse from which one has been removed to the multiplexer 8 via the encoder 7. All of the above-mentioned methods are explained using the case where the pulse interval exceeds the limit at only one place, but it is clear that there is no problem with any number of pulse intervals.

Now, the multiplexer 8 multiplexes the thus obtained multipulse and data regarding the LPC coefficients and transmits the multiplexed data to the decoding side shown in FIG. 1B via a transmission path.

On the decoding side, a demultiplexer 9 performs demultiplexing, and data regarding multipulses is supplied to a decoder 10, and data regarding LPC coefficients is supplied to a decoder 11 for decoding. It is supplied to the LPC synthesizer 12.

The LPC synthesizer 12 has a built-in synthesis filter having a p-order all-pole digital filter configuration, uses multi-pulses as a driving sound source, uses LPC coefficients as filter coefficients, reproduces the input audio signal in digital quantity, and converts it into a D/A converter. 13.

The D/A converter 13 converts the input playback input audio signal into analog, sends it to the LPF 14, cuts off a predetermined high frequency, and then supplies it to the output terminal 1201 as an analog playback input audio signal. .

FIG. 2 is a block diagram showing the configuration of the encoding side in a second embodiment of the present invention.

In the embodiment on the encoding side shown in FIG. 2, only the multipulse searcher 15 and the encoder 16 are
Unlike the content of , other items with the same symbols have the same content, and the decoding side is also almost the same as the content shown in FIG. 1B, so a detailed explanation regarding these will be omitted.

In this second embodiment, the analysis frame is divided into several predetermined small frames, and the multi-pulse is set such that each small frame is always assigned the minimum number of pulses including amplitude 0. The pulse searcher 15 determines a multipulse, and the encoder 16 encodes the multipulse and supplies it to the multiplexer 8, thereby setting limits on adjacent multipulse intervals.

In this case, each small frame provided within the analysis frame does not exceed the pulse interval to be limited. This method is particularly effective when the number of pulses to be set as a multi-pulse is small and the pulse interval is likely to exceed the limit.

FIG. 3 is a block diagram showing the configuration on the encoding side in a third embodiment of the present invention.

The configuration shown in Figure 3 is the multi-pulse searcher 1.
All the parts other than 7 are almost the same as those with the same symbols in FIG. 1A and FIG. Also, the decoding side
Since the content is almost the same as that shown in FIG. B, detailed explanation will be omitted here as well.

The third embodiment shown in FIG. 3 limits the minimum pulse interval when the sound source pulse to be configured as a multi-pulse is processed by the multi-pulse searcher 17 based on the calculation of equation (1). This will be carried out under the following conditions. This method is particularly useful when the number of multipulses to be set is relatively large, so that any loss at the minimum pulse spacing setting will have little effect on the playback audio quality. be.

FIG. 4 is a block diagram showing the configuration on the encoding side in a fourth embodiment of the present invention. In the configuration shown in FIG. 4, the multi-pulse searcher 18 is partially different from that shown in each of the first to third embodiments, and other components with the same symbols are almost the same, and the decoding side is also similar to that shown in the first to third embodiments. Since they are the same as those in the first embodiment, detailed explanations regarding these will be omitted.

In the fourth embodiment, when the multi-pulse searcher 18 searches for sound source pulses to be made into multi-pulses, the maximum pulse width is limited by the following procedure.

In other words, the maximum pulse interval is defined as T-1 slot,
One most likely pulse to be made into a multi-pulse in the range from time slot 1 to time slot T is determined. This maximum likelihood pulse corresponds to the maximum absolute amplitude in the time slot from 1 to T, and this also shows that
It means that g _i (m _i ) is maximum. Next, if the maximum likelihood pulse existence time slot determined in this way is S1, then the time slot S1+
Similarly, one pulse with the maximum likelihood is determined in the range from 1 to S1+T, and the pulse with the maximum likelihood is searched for in the range T over the analysis frame, and these are considered as multi-pulses. In this case, if an apparently extra sound source pulse appears as a result of performing multi-pulse search to the end of the analysis frame, it may be used to set the pulse position to a valid pulse position for the entire analysis frame.

By using any of the first to fourth embodiments described above, it is possible to easily and reliably set the maximum pulse interval for multi-pulses.

The noise weighter 1 used in the first to fourth embodiments described above may be deleted depending on the purpose of encoding or decoding, and the digital filter used as the LPC synthesizer 12 may be a non-polar filter. There is no problem in replacing it with a molded one, etc.

〔Effect of the invention〕

As explained above, according to the present invention, in a multi-pulse type encoding/decoding device that determines pulses in a forward manner, a multi-pulse search is performed using means for limiting the maximum interval between adjacent pulses. This has the effect of realizing a multi-pulse encoding/decoding device that can significantly reduce the number of bits required for pulse interval quantization.

[Brief explanation of drawings]

Figure 1 is a block diagram A showing the configuration of the encoding side in the first embodiment of the present invention, and block diagram B showing the configuration of the decoding side in the first embodiment of the present invention. 3 is a block diagram showing the configuration of the encoding side in the third embodiment of the present invention, and FIG. 4 is a block diagram showing the configuration of the encoding side in the fourth embodiment of the present invention. FIG. 1... Noise weighter, 2... LPC analyzer,
3... Cross-correlation function calculator, 4... Encoder, 5
... Autocorrelation function calculator, 6 ... Multipulse searcher, 7 ... Encoder, 8 ... Multiplexer, 9 ... Demultiplexer, 10, 11 ... Decoder, 12 ... LPC synthesizer , 13...D/A
converter, 14...LPF, 15...multipulse searcher, 16...encoder, 17...multipulse searcher.

Claims (1)

[Claims] 1 Multi-pulses representing the sound source information of the input audio signal are determined through cross-correlation between the input audio signal and the impulse response of a synthesis filter to be placed on the analysis side. The multi-pulse encoding/decoding device searched for is characterized by comprising multi-pulse interval limiting means for limiting the maximum interval between adjacent multi-pulses within the same analysis frame to a preset length. Multi-pulse encoding/decoding device.