JP3046566B2 - Signal analysis method and signal analyzer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a signal analyzing method and a signal analyzing apparatus, and more particularly, to a wide range of applications as a pre-stage processing for a hearing aid signal processing, a speech or music encoding, a signal enhancement, a signal separation, etc. The present invention relates to a simple signal analyzer.

[0002]

2. Description of the Related Art Models for simulating the characteristics of the human auditory peripheral system are roughly classified into the following four types, and the respective features will be described below.

(A) Gamma tone filter This filter is a linear time-invariant filter, and is realized in either a finite-length impulse response (FIR) type or an infinite-length impulse response (IIR) type. This gamma tone filter is now widely used because it is simple and released as public domain software (Patterso
n, RD, Allerhand, M. and Giguere, C. (199
Five)). The function of this gamma-tone filter was proposed because it can well approximate physiological measurement data of cat basilar membrane oscillations (Johannesma, 1972; de Boer and de J
ongh, 1978; Carney and Yin, 1988).

Furthermore, it has been found that this Fourier power spectrum is close to the frequency characteristics of a human auditory filter in the frequency domain measured psychophysically at moderate sound pressure levels. For this reason, it is widely accepted as a filter having a certain background in physiological and psychophysical terms.

However, the gamma tone filter is
It is essentially a time-invariant linear filter, and its power spectrum is almost symmetrical with respect to the center frequency. Therefore, as the sound pressure level increases, the slope of the filter shape on the low frequency side becomes gentler at the center frequency. Psychophysical experimental results (Glasberg and Moore, 1990) and dynamic characteristics when the sound pressure level is changed cannot be simulated as it is.

(B) roex filter In order to explain the sound pressure level dependence that cannot be simulated by the above-described gamma tone filter, a roex filter (Patterson, RD, et al (198) defined only in the frequency domain.
2)) has been used. The roex filter can be well fitted to various psychophysical experiment results, but since the impulse response with time domain is not defined,
It cannot be used as a signal processing model of the auditory peripheral system.

(C) A model for numerically solving the basilar membrane vibration The filter function of the auditory peripheral system is generated as a result of the mechanical basilar membrane vibration. By converting the differential equation of the mechanical vibration into a difference equation, the response of the filter can be solved numerically. In addition, a number of refined models have been proposed by incorporating physiological knowledge (eg, Allen et al (1985), Giguere, C. and
Woodland, PC (1994)). However, they are not well adapted to psychophysical masking data due to their generally large number of parameters and complex responses. Also, it is difficult to control the characteristics due to the dependence of the filter before and after the propagation path. Furthermore, since it is necessary to perform a complicated numerical analysis, it generally takes a long time to calculate.

(D) Gamma chirp filter In order to overcome the disadvantages of (A) to (C),
A "gamma-chirp function" has been proposed that defines the impulse response in the time domain, has asymmetry in the frequency domain, has a small number of parameters, and can be easily adapted to psychophysical data (Irino, 1995, 1996). ). This function has been theoretically derived as an optimal function in the sense of minimum uncertainty in the time-scale representation. By adding only one frequency modulation term to the gamma tone function on the equation, an impulse response with asymmetry can be defined. Further, by making the degree of asymmetry of the gamma chirp filter signal level dependent, it can also serve as a psychophysical background such that it can well match masking data by 12 sets of notch noise methods of three different research institutions. (Irino-Patterson (1996),
Irino and Patterson1997)).

As described above, although the gamma chirp filter has excellent characteristics, the frequency response of the gamma chirp filter is not a simple form given by a time waveform, and therefore, conventionally, only a finite length impulse response (FIR) type is realized. Had not been. In other words, when trying to construct a time-varying filter group as an actual frequency analysis device, an impulse response having the asymmetry that was at the sound pressure level at each sample point interval where each sample point was sufficiently small was obtained, and this was further input waveform. And convolution integration can be realized only by a time-consuming method. If this is further performed for each filter channel, it takes a very long time, and although it depends on the speed of the computer, the sampling rate, the number of channels, and the decimation interval, it takes several tens minutes to several hours to analyze one second of audio data. It turned out to be necessary and far from real time. Although sound analysis alone will eventually yield results, taking into account various fields of application, processing that is infinitely close to real time is required. Also, the actual dynamic characteristics of the auditory peripheral system are constantly changing, and the FIR processing as described above has not been performed. Therefore, the original dynamic characteristics can be completely reflected in the FIR processing as it is. May not occur.

[0010]

Therefore, there is a need for a method of constructing a group of gamma chirp filters that solves these problems. If the gamma chirp filter is composed of a small number of coefficients as an infinite impulse response (IIR) filter group, it can be processed at a speed close to real time despite time deformation. For this reason, the frequency analysis conventionally performed by the fast Fourier transform or the linear filter group can be easily replaced. For this reason, there is an advantage that it can be used for a wide range of signal processing that requires frequency analysis taking human auditory characteristics into consideration as pre-stage processing.

Therefore, a main object of the present invention is to provide a signal analyzing method and a signal analyzing apparatus which can satisfy the above-mentioned requirements.

[0012]

The invention according to claim 1 is
Frequency analysis simulating the basement membrane vibration of the human a possible signal analysis method, the frequency of each frequency band component of the audio signal
Characteristics have almost symmetric characteristics with respect to its center frequency
When analyzed by linear filter unchanged, depending on the magnitude of the output of each frequency band component of the analysis, when the periphery of varying based compensation filter
A coefficient for controlling the asymmetry of the wave number characteristic and a coefficient for the amplitude are determined.

According to a second aspect of the present invention, there is provided a human basement membrane vibration.
A signal analyzer capable of frequency analysis simulating
A linear filter whose frequency characteristic has a substantially symmetric characteristic with respect to its center frequency, and is invariant when an audio signal is input;
As it has asymmetric in the frequency domain, the time the amplitude of the output signal of the linear filter, constituted by a variable system compensation filter when it is possible to change the frequency characteristics.

[0014] In the invention according to claim 3, further according to the amplitude of the output signal of the linear filter, coefficients for determining the coefficients and the coefficient of the amplitude controlling the asymmetry of the frequency characteristic of the compensation filter of varying based upon A decision circuit is provided.

[0015]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Before describing the embodiments of the present invention, the principle of the present invention will be described. The impulse response of the gamma tone filter is given by the following equation (1).

[0016]

(Equation 1)

Here, time t> 0, a is amplitude, and b and n are
Parameter of the comma distribution function, f _cIs the carrier frequency
And φ is the phase, ERB (f_c) Is the equivalent rectangular bandwidth,
ERB (f_c) = 24.7 + 0.108f_c(Hz)
(Glasberg and Moore, 1990).

On the other hand, the impulse response of the gamma chirp filter is expressed by the following equation (2).

[0019]

(Equation 2)

[0020] lnt where time t> 0, f _r is finally converged to the frequency, c is a parameter for determining the degree of asymmetry by a factor of the frequency modulation term represents the natural logarithm. The difference from the gamma tone function is only jclnt, and when c = 0, it clearly matches the gamma tone. This filter was named gamma chirp because it has an FM carrier (chirp signal) with a fractional change in frequency and an envelope of the gamma distribution function. Equivalent rectangular bandwidth varies the signal level, easy in the same manner as in the gamma tone for, as a function of only the frequency f _r.

The amplitude of the Fourier spectrum of the gamma chirp function (complex representation) is obtained as in the following equations (3) and (4).

[0022]

(Equation 3)

When c = 0, it matches the gamma tone.
The part expressed by the fraction of the equation (3) is the characteristic of the gamma tone, and the remaining rightmost term determines the asymmetry with respect to the peak frequency. Expression (3) can be rewritten as expressions (5) and (6), assuming that the amplitude is normalized.

[0024]

(Equation 4)

[0025] Here, | G _T (f) | is the frequency amplitude characteristic of the gamma tone. That is, a gamma tone filter and a filter H _A for converting its characteristics into a gamma chirp
(F) is a tandem connection. Since a method of realizing the gamma tone filter in the IIR type has already been established, a gamma chirp filter can be realized if the characteristics of H _A (f) can be approximately realized with the same number of filter coefficients.

Using the Remez algorithm, F
This characteristic can be realized with a linear phase in the IR type with a high degree of precision, but the number of coefficients increases, and b · c · f _r
It is not practical because it is necessary to calculate a coefficient for each value of or to have a table. Therefore, it is necessary to consider the characteristics in more detail and consider the realization of the IIR type.

FIG. 1 is a diagram showing the frequency characteristic of the above-mentioned equation (6). This characteristic is represented by a solid line, and the following can be said.

(A) When c = 0, the amplitude is 1 at all frequencies, whereas when c is positive, it is a high-pass type, and when c is negative, it is a low-pass type, and the absolute value of c is larger the slope at f _r increases, also spread the range of amplitudes.

(B) It changes monotonically with frequency and does not produce a local maximum value or a local minimum value. (C) The following relationship is established for an appropriate frequency Δf.

[0030]

(Equation 5)

This is different from the characteristics of a relatively commonly used low-pass filter such as a Butterworth type or Chebyshev type, or a characteristic obtained by converting the same into a band-pass / high-pass filter, and cannot be used for an approximate realization. On the other hand, the secondary IIR digital filter used in the present invention is expressed by the following equation.

[0032]

(Equation 6)

(P₁And p_TwoIs a positive constant, f_xIs sampler
Frequency) satisfying the characteristics (a) and (c) in all frequency ranges
be able to. However, in general, local maxima and minima occur.
Therefore, the feature (b) cannot be satisfied.
Therefore, cascade several poles and poles shifted from each other.
To make it as smooth as possible
You. Here, the transfer function H of the compensation filter _C(Z) next
It is realized by the formula.

[0034]

(Equation 7)

The frequency amplitude characteristic of the H _C (z) made with this four-stage shows when normalized by the value at f = f _r by a dotted line in FIG. 1. In Figure 1, the upper and lower center frequency f _r 2ERB
(= 480 Hz) is in good agreement with the characteristic of H _A (f) to be approximated. On the other hand, it can be seen that a deviation occurs as the distance from the center frequency increases. However, basically, this compensation filter is band-controlled by the gamma tone filter of the formula (5).
Therefore, the evaluation is performed by the gamma chirp filter | G _C (f) before and after replacing H _A (f) in Expression (5) with H _A (z).
|

FIG. 2 shows the frequency amplitude characteristics (dotted line) of the gamma chirp filter (solid line) and the gamma tone filter obtained analytically and the frequency amplitude characteristics (dashed line) of the gamma chirp filter produced by applying a compensation filter to the gamma chirp filter.
2, so that characteristics c = -1 are matched better, the amplitude characteristic of the compensation filter is normalized with the frequency _{f = f r + k · p} 3 · c · bERB (f r) at the k-th ing. p ₁ = 1.35−0.19 · | c |, p ₂ =
0.292−0.004 · | c |, p ₃ = 0.058 +
When 0.0018 · | c | is set, the effective error of the two curves in FIG. 2 can be suppressed to about 1 dB. b is 1.
0, 1.35, 1.7, and c are 1, 0, -1, -2,-
3, frequency 200, 400, 800, 1600, 32
In the case of 90 combinations of 00 and 6400 Hz, the average effective error is 0.6 dB, and the
There were only examples. In this way, the characteristics can be matched very well. Furthermore, optimization is performed, and
There is a possibility that it can be further improved by changing the way of taking the coefficient of the expression (16) from the expression (4).

As described above, it has been found that by combining a linear time-invariant gamma tone filter and a compensation filter, the characteristics of gamma chirp can be accurately realized.
Since this compensation filter is composed of an IIR type filter, the overall operation time can be extremely short if a well-known IIR type gamma tone filter is combined. A gamma chirp filter having such a configuration is provided by IIR
We call it the shape gamma chirp filter. here,
The characteristics of the compensating filter need not be fixed, and can be changed over time.

Hereinafter, an embodiment based on the above principle will be described. FIG. 3 is a block diagram showing one embodiment of the present invention. In FIG. 3, an audio signal or the like is input to a time-invariant linear frequency analysis filter group 2 via an input unit 1. The time-invariant linear frequency analysis filter group 2 is composed of a plurality of band analysis filters each having a characteristic frequency, and an input audio signal is analyzed into each frequency band component. This bandpass filter corresponds to the aforementioned gamma tone filter. Time-invariant linear frequency analysis filter group 2 of the output signal 3 is time-varying system compensation filter group 4 and a coefficient determining circuit group 7
And given to. Is intended the coefficient determining circuit group 7 which determines the asymmetry parameter c and the amplitude parameter a for each channel of varying based compensation filter group 4 when, and outputs the determined parameter time to varying system compensation filter group 4.

FIG. 4 is a block diagram of the coefficient determining circuit group 7 shown in FIG. In FIG. 4, a coefficient determining circuit group 7 includes a rectifier circuit 9, a leaky integrator 10, and a coefficient determining circuit 11 corresponding to each channel, and is a time-invariant linear frequency analysis filter group which is a gamma tone filter group. Output 6 of 2
Is rectified by the rectifier circuit 9 for each channel, and then integrated by the leaky integrator 10 to average the output. This is weighted and added from adjacent channels to obtain a coefficient determining circuit 11.
Is input to The coefficient determination circuit 11 determines the parameters c and a from the given activity, and can reasonably determine the parameters c and a from the result of the psychophysical experiment. The output 8 of the coefficient determining circuit group 7 as described above, when given to the variable system compensation filter group 4, the final output 5 equivalent to that over gammachirp filter group of variable system time to the input The output is obtained.

[0040]

As described above, according to the present invention, the power spectrum has a characteristic substantially symmetrical with respect to the center frequency thereof, and the output of the linear filter which is invariable when the audio signal is input is used as the compensation filter. Since the time and frequency characteristics are given to change the dynamic characteristics of the basilar membrane of the auditory peripheral system, physiological knowledge about the dynamic characteristics of the basement membrane of the auditory peripheral system can be given and simulated more clearly. This is important for signal processing taking human processing systems into account. In addition, processing that is close to real-time processing can be performed with the temporally deformed auditory filter group. That is, in the signal processing in which the auditory filter group is used as an input stage, the processing speed can be improved by replacing the input stage.

[Brief description of the drawings]

FIG. 1 is a diagram showing an amplitude frequency characteristic of a gamma tone filter and an amplitude frequency characteristic of a compensation filter.

FIG. 2 is a diagram showing amplitude frequency characteristics of a filter according to the present invention.

FIG. 3 is a block diagram showing an embodiment of the present invention.

FIG. 4 is a block diagram illustrating an example of a coefficient determination circuit illustrated in FIG. 3;

[Explanation of symbols]

1 input, 2 time invariant linear frequency analysis filter group, 4
Time-varying system compensation filter group, 7 coefficient determining circuit group, 9 rectifier circuit, 10 leaky integrator, 11 coefficient determining circuit.

──────────────────────────────────────────────────続 き Continuation of the front page (51) Int.Cl. ⁷ Identification code FIG10L 9/10 (56) References JP-A-1-191510 (JP, A) JP-A-11-119797 (JP, A) Patent 2588004 (JP, B2) JP 7-27398 (JP, B2) The Journal of the Acoustic Society of America, Vol. 1, January 1997, "A time-domain, level-dependant au directory filter: The gammachirp", p. 412-419, The Journal of the Acoustic Society of America, Vol. 4, Pt. 1, April 1996, "Temporal asymmetry in the audiory system", p. 2316-2331 The Journal of the Acoustic Society of America, Vol. 5, November 1998, "Modeling temporal assembly in the auditory system", p. 2967-2979 Processes of 1988, IEEE International Conference on Acoustics, Speech and Signal Processing, Vol. 6, "A time-varying, analysis / synthesis audition filter bank using the gammachirp" p. 3653-3656 Proceedings of 1996 IEEE International Conference on Acoustics, Speech and Signal Processing, Vol. 2, "A gammach irp'function as an optimal audition filter with the mellin transform", p. 981-984, Materials of the Acoustical Society of Japan, H-98-98, "Time-Varying Analysis Synthetic Gamma Chirp Auditory Filter Bank" (Announced on September 18, 1998) Material of the Acoustical Society of Japan H-97-69 "Efficient Configuration of Gamma Chirp Filter and Filter Bank" (announced on October 24, 1997) Material of the Acoustical Society of Japan H-96-73, "Gamma chirp as a level-dependent auditory filter" (published October 18, 1996) Proceedings of the Acoustical Society of Japan Fall Meeting 1997-I ▼ 1-3-16 “IIR Filter” Realization of a gamma chirp filter by ”p. 421-422 (published September 17, 1997) Proceedings of the Acoustical Society of Japan 1998 Spring Meeting, ▲ I ▼ 1-8-2 “Time-Varying System Analysis-Synthesis Auditory Model Using Gunma Chirp Filter Bank” p . 413-414 (published March 17, 1998) Proceedings of the Acoustical Society of Japan, Spring Meeting, 1998 [I] 1-8-3 “Asymmetric control method in Gunma chirp filter bank” p. 415-416 (issued on March 17, 1998) (58) Fields investigated (Int.Cl. ⁷ , DB name) G10L 11/00 H03H 17/02 601 H03H 17/04 615 H03H 17/04 633 H03M 7 / 30 INSPEC (DIALOG) JICST file (JOIS) WPI (DIALOG)

Claims (3)

(57) [Claims] 1. A signal analysis method capable of performing frequency analysis simulating human basilar membrane vibration, wherein a frequency characteristic of each frequency band component of an audio signal
Time-invariant linear filter with nearly symmetrical characteristics
Analyzed by filter, according to the magnitude of the output of each frequency band component analyzed, and determines the coefficients and the amplitude coefficient of controlling the asymmetry of the frequency characteristics of the variable based compensation filter, the signal analysis Method. 2. A signal analyzer capable of frequency analysis simulating human basilar membrane vibration , wherein the frequency characteristic has a substantially symmetrical characteristic with respect to a center frequency thereof, and is invariable when an audio signal is input. and linear filter, that has an asymmetry in the frequency domain, the time the amplitude of the output signal of the linear filter, and a varying system compensation filter when it is possible to change the frequency characteristic, the signal analyzer. 3. Further, according to the amplitude of the output signal of the linear filter, the frequency characteristic of the compensation filter of the time-varying system
3. The signal analyzer according to claim 2, further comprising a coefficient determination circuit for determining a coefficient for controlling the asymmetry of the signal and a coefficient for the amplitude.