JP3163206B2 - Acoustic signal coding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio signal encoding apparatus for compressing and encoding an audio signal or an audio signal for communication or storage.

[0002]

2. Description of the Related Art As a first prior art, there is a technique of spectrally shaping quantization noise generated by encoding when compressing and encoding an audio signal by utilizing an auditory masking characteristic. One example is "A New Model of LP
C Excitation for Producing Natural-Sounding Speech
at Low Bit Rates ", BSAtal and JR Remde, IEEEI
nt. Conf.on Acoustics, Speech and Signal Processin
g, pp. 614-617, 1982.

[0003] This uses a linear prediction coefficient obtained by performing a linear prediction analysis on an audio signal to filter a quantization error waveform with a filter having a transfer characteristic expressed by the following equation (1). This is a method of performing an encoding process so as to minimize energy.

[0004]

(Equation 1)

In the above equation (1), a _k represents a _k-th linear prediction coefficient, p represents a prediction order, and β and γ represent constants satisfying 0 ≦ γ ≦ β ≦ 1, respectively.

A codebook-excited linear predictive coding (Code-Ex), which is a speech coding method using this auditory weighting filter,
cited Linear Predictive Coding. Hereinafter referred to as CELP) is, for example, “Code-Excited Linear Prediction
(CELP): High-Quality Speech at Very Low Bit Rates ",
MRSchroeder and BSAtal, IEEE Int.Conf.on Acoust
ics, Speech and Signal Processing, pp. 937-940, 1985. FIG. 4 is a block diagram showing the configuration.

In FIG. 4, 1 / A (z) is given by the following equation (2).
Is a speech linear prediction synthesis filter represented by

[0008]

(Equation 2)

In the above equation (1), γ = 0.8, β = 1
, And the audio linear prediction synthesis filter represented by the above equation (2) is synthesized with the perceptual weighting filter, the result is simplified as in equation (3).

In this case, the block diagram of FIG. 4 is changed to the configuration shown in FIG.

[0011]

(Equation 3)

In the prior art described above, the auditory weighting filter represents the auditory masking characteristic as a characteristic that is very easily approximated.

As a second conventional technique, there is a technique used in compression coding of an audio signal. This method is
The auditory masking characteristic is more actively used than the first prior art.

FIG. 6 shows an operation sequence of an audio signal encoding unit used in MPEG. One example is
"High Efficiency Coding of Audio Signal-MPEG Audio Coding System" Goto, Journal of the Acoustical Society of Japan, Vol. 966
-969, 1991.

In the upper right part of the flow of FIG. 6, a power spectrum is obtained from an input signal using FFT, and an auditory masking characteristic is calculated from information on the power spectrum.
In MPEG Layers 1 and 2, band division coding is basically used, and coded bits for each band are determined from information on masking characteristics and the like.

As a third conventional technique, there is a technique that combines the first and second techniques. In this method, an auditory masking characteristic is obtained from power spectrum information, and an encoding process is performed using an auditory weighting filter having the inverse characteristic to minimize the energy of the quantization error waveform. One example is "Some Experiments in Perceptual Maskinig ofQua
ntizing Noise in Analysis-By-Synthesis Speech Code
rs ", R.Drogo De Iacovo and R.Montagna, EUROSPEECH, p
pp. 825-828, 1991.

In this method, a minimum phase finite impulse response filter (hereinafter, referred to as an FIR filter) having a power spectrum characteristic of an auditory masking characteristic is designed using a Hilbert transform technique, and its inverse filter is weighted by an auditory weight. Used as a filter.

[0018]

However, the characteristics of the auditory weighting filter in the first prior art described above are different from human auditory masking characteristics because they are obtained by simple approximation, and the quantization noise However, there was a problem that it was not possible to conceal sufficiently.

In the second prior art, the masking characteristic is obtained in accordance with the model of the human auditory masking characteristic. However, the band division coding is finally used, and the addition of bit allocation and the like is performed. There is also a problem that information is required and the compression ratio cannot be sufficiently reduced.

Further, in the third prior art described above, by coping with the above two problems, taking into account the auditory masking characteristics, and using an auditory weighting filter, an encoding system with a high compression rate can be realized. . However, since the auditory weighting filter is composed of the FIR filter, it is inferior to the infinite impulse response filter (hereinafter referred to as IIR filter) from the viewpoint of approximating the amplitude frequency characteristic with the same filter order, and the first conventional technique. However, there is a problem that it is difficult to simplify the processing by the synthesis processing of the auditory weighting filter and the linear predictive synthesis filter of the voice as described in (1).

An object of the present invention is to provide an audio signal encoding apparatus capable of sufficiently concealing quantization noise, sufficiently reducing the compression ratio, and simplifying the entire processing, in view of the above-mentioned problems in the prior art. To provide.

[0022]

SUMMARY OF THE INVENTION It is an object of the present invention to obtain a power spectrum of an audio signal, obtain an audio masking spectrum characteristic, a first filtering means having an inverse power spectrum characteristic of the audio signal, A second filtering unit having a spectral characteristic obtained by dividing a power spectral characteristic of the signal by an auditory masking spectral characteristic, and achieved by an audio signal encoding device that performs an auditory weighting process by the first filtering unit and the second filtering unit. Is done.

An audio signal encoding apparatus according to the present invention is configured to include an inverse Fourier transform unit for obtaining an autocorrelation sequence from a power spectrum of an audio signal, and a unit for calculating a coefficient of a second filtering unit from the autocorrelation sequence. You may.

The audio signal encoding apparatus according to the present invention comprises means for obtaining a logarithmic power spectrum, means for obtaining a cepstrum from the logarithmic power spectrum by inverse Fourier transform, and means for calculating a coefficient of the second filtering means from the cepstrum. It may be configured as follows.

[0025]

In the audio signal encoding apparatus according to the present invention, the power spectrum of the audio signal is determined, the auditory masking spectrum characteristic is determined, the first filtering means has the inverse power spectrum characteristic of the audio signal, and the second filtering means has the audio power spectrum characteristic. It has a spectral characteristic obtained by dividing the power spectral characteristic of the signal by the auditory masking spectral characteristic, and performs an auditory weighting process by the first filtering means and the second filtering means.

In the audio signal encoding apparatus according to the present invention, the inverse Fourier transform means obtains an autocorrelation sequence from the power spectrum of the audio signal, and calculates a coefficient of the second filtering means from the autocorrelation sequence.

In the audio signal encoding apparatus according to the present invention, a logarithmic power spectrum is obtained, a cepstrum is obtained from the logarithmic power spectrum by inverse Fourier transform, and a coefficient of the second filtering means is calculated from the cepstrum.

[0028]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing an embodiment of an audio signal encoding apparatus according to the present invention.

FIG. 1 is a block diagram showing the configuration of a first embodiment of an audio signal encoding apparatus according to the present invention, and shows an example using a CELP system.

The audio signal encoding apparatus shown in FIG. 1 is connected to an input terminal 105 of an audio signal and an LP for performing linear prediction analysis (hereinafter referred to as LPC analysis) on the audio signal.
A power spectrum calculation unit 111 connected to the C analysis unit 110 and the LPC analysis unit 110 to calculate the power spectrum P (ω) of the signal from the LPC analysis result. A masking characteristic calculator 112 for calculating a masking characteristic M (ω), a power spectrum calculator 111, and a divider 11 connected to the masking characteristic calculator 112 for dividing the power spectrum of the signal by the masking characteristic.
3. The input signal is connected to the IIR filter coefficient calculation unit 114, the input terminal 105, and the LPC analysis unit 110 which are connected to the divider 113 and obtain the IIR filter coefficient from the spectrum ratio characteristic obtained by the divider 113. FI as first filtering means for auditory weighting
The IIR filter 109, which is connected to the R filter 107, the FIR filter 107, and the IIR filter coefficient calculation unit 114 and is a part of the second filtering means for perceptually weighting the input signal, an excitation codebook for CELP speech coding ( Codebook) 101, codebook 101
, An amplification unit 102 for amplifying the excitation signal, a pitch component synthesis filter 103 connected to the amplification unit 102 for synthesizing pitch components, and a pitch component synthesis filter 1
IIR filter 104, IIR filter 1 which is connected to the IIR filter coefficient calculating unit 114 and is another part of the second filtering means having the characteristic of combining the speech spectrum synthesis filter and the auditory weighting filter.
The subtraction unit 106 is connected to the subtraction unit 106 and the subtraction unit 106. The subtraction unit 106 is connected to the subtraction unit 106 and is connected to the subtraction unit 106 to reduce the energy of the difference waveform. It is configured by an energy minimizing unit 108 for setting the optimization parameter.

In this embodiment, the configuration of the auditory weighting filter is different from that of the prior art shown in FIG. The following description focuses on the configuration of the auditory weighting filter.

The signal input from the input terminal 105 is subjected to a segmentation process for every certain time length. This is called a frame. The signal of one frame is output to the LPC analysis unit 11
At 0, a linear prediction coefficient is calculated. This linear prediction coefficient is set as a coefficient of the FIR filter 107 which is a numerator of the auditory weighting filter shown in the above equation (1) (hereafter, β = 1 in the above equation (1)). The amplitude transfer characteristic is calculated by the power spectrum calculation unit 111 from the calculated linear prediction coefficient. From the transfer characteristic expressed by the above equation (2), a power spectrum is calculated by the following equation (4).

[0033]

(Equation 4)

In the above equation (4), when ω = 2πFs, F
s is the sampling frequency.

In the above description, the power spectrum of the input signal is calculated from the result of the LPC analysis. However, the input signal may be calculated by performing a Fourier transform. In this case, since the frequency resolution is required to be higher than the LPC spectrum, the masking characteristic can be calculated more accurately.

The masking characteristic calculator 112 calculates a masking spectrum characteristic from the power spectrum of the input signal. The outline of this processing procedure is to decompose the power spectrum for each critical auditory bandwidth, calculate the masking curve of the quantization noise due to the input signal for each critical band, and obtain the minimum audible value and time over the entire signal band. A masking curve M (ω) is calculated in consideration of masking at the axis and the like. Various calculations of a masking curve have been proposed, and one example is “Estimation of Perceptual Entropy Using Noise”.
Masking Criteria ", JD Johnston, IEEE Int. Conf.on A
coustics, Speech and Signal Processing, pp. 2524-252
There are 7,1988.

The auditory weighting filter is designed so that the quantization noise due to the encoding is shaped according to the shape of the masking curve. That is, the power spectrum of the auditory weighting filter needs to have the inverse characteristic of the masking spectrum. Here, a filter F (z) that satisfies the relationship shown in the following equation (5) is considered.

[0038]

(Equation 5)

The amplitude transfer characteristic of the filter F (z) can be considered as a filter having a transfer characteristic obtained by dividing the power spectrum P (ω) of the input signal by the masking spectrum M (ω).

When this filter F (z) is realized by an all-pole IIR filter, the auditory weighting filter W
When (z) and the speech synthesis filter 1 / A (z) are synthesized,
It can be simplified as in the following equation (6).

[0041]

(Equation 6)

In order to perform the above operation, the dividing unit 113 sets P
(Ω) / M (ω) is obtained, and the IIR filter coefficient calculation unit 1
At 14, an IIR filter coefficient is calculated from the power spectrum represented by P (ω) / M (ω).

Here, the transfer function of the filter F (z) is shown in equation (7).

[0044]

(Equation 7)

In equation (7), q is the order of the IIR filter and does not need to match the linear prediction order of the speech. f _k is a k-th order coefficient of the IIR filter calculated by the IIR filter coefficient calculation unit 114.

In the above processing, the FIR filter 107 having the transfer function of A (z) in which the linear prediction coefficient obtained by the LPC analysis unit 110 is set and the IIR filter described above
Aurally weight the input signal. The IIR filter 10 having the same coefficient as the IIR filter 109 is set.
4 obtains an auditory weighted reproduction signal. The subsequent encoding process is the same as that of a general CELP encoding method, and will be briefly described so that the error energy between the input signal weighted perceptually and the reproduced signal weighted perceptually is minimized. , And determine the encoding parameters.

Next, a method of calculating an IIR filter coefficient from the power spectrum P (ω) and the masking spectrum M (ω) will be described.

FIG. 2 shows a processing procedure for calculating an IIR filter coefficient from the power spectrum P (ω) and the masking spectrum M (ω) by solving the inverse Fourier transform and the normal equation.

Hereinafter, description will be made with reference to FIG.

First, a power spectrum is defined as shown in the following equation (8).

[0051]

(Equation 8)

Since the power spectrum S (ω) and the autocorrelation function R (τ) have a relationship as shown in the following equation (9), the FFT method and the like can be performed in the range of τ = 0 to q. Is used to calculate the autocorrelation sequence.

[0053]

(Equation 9)

Next, the conversion from the autocorrelation coefficient to the IIR filter coefficient is obtained by solving the normal equation of Expression (10) as generally used in the linear prediction analysis of speech.

[0055]

(Equation 10)

In equation (10), (...) ^T represents a matrix transpose operation.

With the above operation, the coefficients of the IIR filter are calculated.

FIG. 3 shows another method of calculating an IIR filter coefficient using a cepstrum by homomorphic processing.

Here, since the operation of dividing the power spectrum P (ω) by the masking spectrum M (ω) is performed in the logarithmic domain, the processing shown in equation (11) is performed. This is because in FIG. 3, P (ω) and M (ω) are logarithmic calculation units 301 and 301, respectively.
This corresponds to logarithmization at 302 and subtraction at arithmetic unit 303.

[0060]

[Equation 11]

This Log S (ω) is calculated by the inverse FFT operation unit 3
When the inverse Fourier transform is performed at step 04, the cepstrum c _n is calculated according to equation (12) (“basic of speech information processing”).
Saito, Nakata, Ohmsha, pp. 99-103).

[0062]

(Equation 12)

Since the lower order part of the cepstrum c _n represents the spectral structure, the cepstrum window (eg, w _n
= 1: n = 1~q, w n = 0: to windowing with n> q).
From the cepstrum c _n thus obtained, the coefficients of the IIR filter are calculated by equation (13).

[0064]

(Equation 13)

However, in the equation (13), k is k = 1.
To q.

Although the above description has been made with reference to the CELP system, the present invention can be easily applied to a system having an auditory weighting filter such as multi-pulse coding as a component.

By changing the part related to the coding of the coding apparatus, the decoding apparatus can be realized without changing other parts.

[0068]

According to the present invention, there is provided an audio signal encoding apparatus comprising: means for obtaining a power spectrum of an audio signal; means for obtaining an auditory masking spectrum characteristic; first filtering means having an inverse power spectrum characteristic of the audio signal; A second filtering unit having a spectral characteristic obtained by dividing a power spectrum characteristic of the signal by an auditory masking spectral characteristic, and performing an auditory weighting process by the first filtering unit and the second filtering unit. Noise noise can be shaped by an auditory weighting filter, and the auditory masking of human auditory characteristics can be used to make the noise less audible and improve the reproduction sound quality.
Further, by synthesizing the auditory weighting filter with the speech linear prediction synthesis filter, the simplification can be performed, and the amount of encoding operation can be reduced.

The audio signal encoding apparatus of the present invention can effectively calculate the coefficients of the inverse Fourier transform for obtaining the autocorrelation sequence from the power spectrum of the audio signal and the second filtering means from the autocorrelation sequence.

The acoustic signal encoding apparatus according to the present invention uses a means for obtaining a logarithmic power spectrum and a means for obtaining a cepstrum by inverse Fourier transform from a logarithmic power spectrum to effectively calculate the coefficient of the second filtering means from the cepstrum. Can be.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of an embodiment of an audio signal encoding device according to the present invention.

FIG. 2 is a flowchart for explaining an example of filter coefficient calculation in the audio signal encoding device of the present invention.

FIG. 3 is a block diagram for explaining another example of filter coefficient calculation in the audio signal encoding device of the present invention.

FIG. 4 is a block diagram illustrating a conventional CELP speech coding scheme.

FIG. 5 is a block diagram for explaining simplification of an auditory weighting filter process of a conventional CELP speech coding scheme.

FIG. 6 is a flowchart for explaining a conventional encoding method in consideration of auditory masking.

[Explanation of symbols]

 101 CELP excitation codebook (codebook) 102 Multiplication unit 103 Pitch component synthesis filter 104, 109 All-pole IIR filter 105 Input terminal 106 Subtraction unit 107 FIR filter 108 Error energy minimization unit 110 Linear prediction analysis unit 111 Power spectrum calculation Unit 112 masking characteristic calculation unit 113 spectrum ratio calculation unit 114 all-pole IIR filter calculation unit 301, 302 logarithmic calculation unit 303 subtraction unit 304 inverse Fourier transform unit 305 cepstrum windowing unit 306 conversion unit from cepstrum to prediction coefficient

Claims (3)

(57) [Claims] 1. A means for obtaining a power spectrum of an acoustic signal, a means for obtaining an auditory masking spectrum characteristic,
A first filtering unit having an inverse power spectrum characteristic of the audio signal; and a second filtering unit having a spectrum characteristic obtained by dividing a power spectrum characteristic of the audio signal by the auditory masking spectrum characteristic.
An audio signal encoding device, wherein an audio weighting process is performed by the first filtering means and the second filtering means. 2. The apparatus according to claim 1, further comprising: an inverse Fourier transform unit for obtaining an autocorrelation sequence from a power spectrum of the acoustic signal; and a unit for calculating a coefficient of the second filtering unit from the autocorrelation sequence. 2. The audio signal encoding device according to claim 1. 3. A means for obtaining a logarithmic power spectrum,
2. The audio signal encoding according to claim 1, further comprising: means for obtaining a cepstrum from the logarithmic power spectrum by inverse Fourier transform; and means for calculating a coefficient of the second filtering means from the cepstrum. apparatus.