JPS6310395B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to an apparatus for analyzing waveforms, and more particularly to an apparatus for analyzing temporal and frequency characteristics of waveforms and extracting the characteristics. The target waveforms are all kinds of waveforms, including impulse type, damped sinusoidal type, approximately periodic type, periodic and stationary type, and combinations thereof.

Conventional configurations and their problems Until now, the following methods have been used to analyze waveforms in the fields of acoustics, communications, etc.

(1) Method using band filter group (2) Method using heterodyne detection using one band filter and variable reference frequency oscillator (3) Fast Fourier transform (FET) method (4) Method using linear prediction (LPC) method (5) Method using autocorrelation function (6) Method using zero-crossing waves These methods do not have sufficient analytical power for highly non-stationary waveforms. One such waveform is audio.

In communication systems, a spectrum analyzer is sufficient for frequency analysis of signal waveforms. In communication systems, a non-stationary modulated wave is applied to a stationary signal called a carrier wave. Generally, frequency analysis is performed at the stage of designing and manufacturing a communication system, and at this time, the modulated wave is also a stationary signal, so that the signal consists of a carrier wave and stationary sideband waves. Since this is a stationary signal, any of the above-mentioned analysis methods will work, and in particular, a commercially available spectrum analyzer can perform measurements with sufficient accuracy.

In recent years, FET has been widely used to analyze the waveforms of musical instrument sounds, which are described in terms of a fundamental tone and its harmonics. Depending on the type of instrument, it has given satisfactory results for steady-state sounds, but for percussion instruments, etc. Good results cannot be obtained for non-stationary sounds. Obtaining good results means that the characteristics of the sound can be sufficiently described with a few parameters.

The above method is also used to analyze voice waveforms. However, voice waveforms do not have a stationary carrier wave like communication signals. It is also not a quasi-periodic waveform like string instruments or wind instruments. Speech consists of a consonant part and a vowel part, but the consonant part is completely non-stationary, and even the vowel part consists of a series of transient waveforms based on the impulse waveforms emitted by the vocal bands. Since the changes in that column are large, analysis that treats it as a stationary column cannot analyze rapid changes or subtle changes.

Purpose of the Invention The purpose of the present invention is to analyze unsteady waveforms such as the above-mentioned voice waveforms and percussion instruments with sufficient accuracy, and determine parameters such as the impulsiveness of the waveform, the vibration frequency and damping rate of the damped sine wave, etc. from the analysis results. The present invention aims to provide a waveform analyzer that derives the waveform characteristics and obtains the characteristics of the waveform.

Configuration of the Invention The waveform analyzer of the present invention usually has a bandwidth of Δ(i),
A plurality of n single resonance characteristics with a center frequency of ₀ (i) and i = 1 to n arranged with mutual center frequency intervals of about the passband width Δ(i) or smaller than that. A first block including a frequency selection filter, a second block that measures the frequency corresponding to each period from the output waveform of each of the frequency selection filters, and quantizes the measured frequency of the output waveform of each of the frequency selection filters. a third block that detects the amplitude of the output waveform of each of the frequency selection filters, and a fourth block that detects the amplitude of each output waveform of the fourth block to the frequency channel to which the waveform belongs. It consists of a channel register block for registering, a self-channel register block for registering the center frequency response of each filter corresponding to the frequency channel, and a separately excited channel register block for registering other responses. This allows frequency analysis of input signals with high accuracy.

DESCRIPTION OF EMBODIMENTS Hereinafter, embodiments of the present invention will be described with reference to the drawings. Although the embodiment will be described here with reference to audio waveforms, the present invention is not limited to audio waveforms only. FIG. 1 shows a block diagram of an embodiment of the invention. In FIG. 1, a signal is input from an input 101. Each of the filters a ₁ , a ₂ , a ₃ ...a _j ...a ₂₅₀ of the filter block 102 is a filter having a frequency characteristic as shown in FIG. The most typical transfer function expression is in Laplace transform form: Gj (s) = 2ζ (TS) / 1 + 2ζ (TS) + (TS) ² × [2TS / (1 + TS) ² ] ² ... (1) (1) In the formula, the value of ζ is 1 or less. At this time, the first term on the right side exhibits bandpass characteristics without resonance. ζ has the following relationship with so-called Q.

ζ=1/2Q...(2) Q=1/2ζ...(3) When the analysis target is audio, the Q value is selected from 1 to 10 for 1KHz or less, and from 3 to 100 for 1KHz or more. These filters may be analog filters or digital filters. When using a digital filter, an analog-to-digital converter is used for the input, and the Z-conversion format corresponding to equation (1) is also used for the filter. The filter group is arranged densely so that ₀ (i+1) - ₀ (i)≦Δ(i), where the center frequency of the filter is ₀ (i) and the band width is Δ(i). Although the center frequencies are arranged in the form of an arithmetic progression in this embodiment, a geometric progression arrangement or a combination thereof may be used depending on the case. Each filter output of the filter block 102 becomes a signal waveform that changes dynamically over time. This output waveform is adjusted to saturation value ±1 by waveform shaping block 103.
Waveform shaping section b ₁ , b ₂ , b ₃ ... b _j ... with infinite gain
...It is shaped with b ₂₅₀ and becomes a square wave. This aspect is the third
As shown in the figure. The rectangular wave output from the waveform shaping block 103 is input to the period measuring section d ₁ , _{d 2} , d _{3 .} . . d _j . The time interval from start to rise is measured. When signal processing is performed using a digital method, the accuracy of the zero-crossing time may be insufficient due to the sampling time interval, but in this case, this is compensated for by interpolation. Each measured value is a single value or an average value of an appropriate number of measured values is sent to the channel number discriminating section e ₁ , e ₂ , e ₃ , ... e _j ... e ₂₅₀ of the next channel number discriminating section 106 as a cycle measurement block. It is input as the output of 105. A channel refers to a sufficiently fine divided section of the frequency domain to be measured as shown in FIG. It is. Further, the channels are configured so that the center frequency ₀ (i) is not at the boundary but at the center of a certain channel. ₀
The channel to which (i) belongs is named the center channel. Also, the channels other than the central channel are called side channel channels. In this example, the case where zero crossing time is adopted is shown as an example of period measurement, but in addition, FET,
Techniques such as autocorrelation functions are used. A channel number determination block 106 determines to which of the above-mentioned channels, that is, each channel shown in FIG. 4, the output of the period measurement block 105 belongs.

On the other hand, the output of each filter of the filter group 102 is transmitted to each amplitude detection section C ₁ of the amplitude detection block 104,
C ₂ , C _{3 .} . . C _j . That value continues to be held for the duration of the cycle. The value is then input to the registration block 107 and sent to the registration section g ₁ ,
g ₂ , g ₃ ... g _j ... registered in g ₂₅₀ . Here, the registration section is composed of a channel register for registering a calculation result value and an arithmetic circuit for calculating (for example, adding) the value already in the channel register and the input value. Furthermore, registration refers to the arithmetic processing and registration processing performed by the registration section. More advanced registration, such as vector addition that takes phase into consideration, is also used.
Registration block 107 consists of various channel register blocks 108 and 109. Channel register block 108 consists of a center channel register group and a side band channel register group. In the channel registration section, the channel number to be registered has already been input from the channel number determination block 106. Therefore, for example, the peak value of the filter output is added to the channel register at the frequency corresponding to the zero crossing time of that waveform. Furthermore, the channel register block 109 corresponds to the central channel register and has a self-channel register block 11.
1, and a separately excited channel register block 112 corresponding to all channel registers. When the output of the period measurement block corresponding to a certain filter falls within the frequency range of its own center channel, the amplitude detection section output is registered in the corresponding self-channel register. Further, when the output of the period measurement block enters the frequency region of the sideband channel or the center channel of another filter, the amplitude detection section output is registered in the corresponding separately excited channel register. Each register in FIG. 4 is reset at appropriate intervals. For example, in the case of audio, the time point at which the peak value of each filter output is updated, the pitch period, or the frame period of several milliseconds to several tens of milliseconds are used. Channel register block 109
The self-channel sum register 114 located within calculates and holds the sum or rms sum of all self-channel registers 111, and also contains the separately excited channel sum register 11.
5 calculates and has the sum or rms sum of all independently excited channel registers 112.

The total channel sum register 113 calculates and holds the sum or rms sum of the value of the self-channel sum register 114 and the value of the separately excited channel sum register 115.

The configuration of the present invention as described above represents the basic concept, and the means for realizing each block is not limited to this embodiment. for example,
Needless to say, this system can implement all blocks using digital computer software after converting the input signal from analog to digital.

The gist of the present invention is to (i) provide a group of low-Q single-tuned filters that are arranged fairly densely, (ii) obtain the frequency by measuring each period of the output wave of the filter group, and (iii) obtain the frequency. (iv)
The main idea is to increase the frequency resolution power by registering each filter output to the assigned channel, and by optimizing the analysis frequency channel array, the self-excited vibration and externally excited vibration of each filter can be reduced. The characteristics of the signal waveform are extracted by separating the non-oscillatory impulse signal and the oscillatory signal so that they can be identified and extracting the characteristics. The operation in such a configuration is described in three typical examples.
Let's explain with one example.

(1) When the input is an ideal impulse.

Since the past value is zero, the initial value of the output of the filter group is zero, and each filter in the filter block 102 performs its own impulse response starting from the time of impulse input. The impulse response is almost a damped oscillation at the center frequency of each filter. Therefore, the resulting channel whose waveform is shaped by the waveform shaping block 103, whose period is measured by the period measuring block 105, and which is determined by the channel number determining block 106 becomes the center channel of each filter, that is, the self channel. Therefore, for isopulses, the n self-excitation channels output their outputs all at once, and the separately excited channels do not output their outputs. Therefore, self-channel sum register 11
4 gives a large value, separately excited channel sum register 1
The value of 15 is zero.

(2) When the input is a single continuous sine wave.

Since each filter in filter block 102 is linear, if the input is a sine wave, the output will also be a sine wave with the same frequency. That is, all filter outputs are sine waves of the same frequency. Therefore, as a result of waveform shaping, period measurement, and channel number discrimination, each channel discrimination block all specifies the same channel number. Therefore, for example, if the current frequency is 44 Hz, the channel number will be 13, and only the 13th channel of the channel register will show a non-zero value, and the others will be zero. That is, a sharp single line spectrum is obtained.

(3) For vowels.

Vowels are formed by exciting the resonant part of the vocal tract by vocal cord vibration pulses. When the waveform of a certain vowel with a low pitch frequency, such as "a", is analyzed using this system, an output appears in the self-channel in the region of several hundred Hz, and in the region above several hundred Hz, it corresponds to the formant frequency due to vocal tract resonance. Output appears on the externally excited channel. In other words, the pitch periodic component of the vocal cords and the vocal tract resonance component are well separated, and the vocal tract resonance component is similar to the case of a sustained sine wave.
The channel is narrowed down and formants are clearly detected.

Effects of the Invention Waveform components are expressed as impulses, sustained sine waves, damped sine waves, white noise, etc., but in conventional methods, waveform components are It has been difficult to measure short-time phenomena such as damped sine waves with high frequency accuracy. This is called the uncertainty principle between time and frequency. The present invention has achieved considerable improvements. It is especially suitable for voice analysis. Although it has been said that the first to third vibrations of the vocal cords during vowel production are important, the information structure has not been elucidated due to the lack of analysis methods, but according to the present invention, clearly explained.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing one embodiment of the present invention;
Figure 2 is a diagram showing an example of the frequency characteristics of a filter.
Figure 3 is a diagram showing the processing process in the waveform shaping section, Figure 4
The figure is a diagram showing an example of registers for various channels. 102... Filter block, 103... Waveform shaping block, 104... Amplitude detection block, 1
05...Period measurement block, 106...Channel number discrimination block, 107...Registration block, 1
08...Channel register, 111...Self channel register, 112...Separately excited channel register.

Claims (1)

[Claims] 1 The passband width is Δ(i), the center frequency is ₀ (i),
a first block including a plurality of frequency selection filters having a single resonance characteristic, i=1 to n, arranged with mutual center frequency intervals of about the passband width Δ(i) or smaller; A second block that measures frequencies corresponding to each cycle from the output waveforms of the frequency selection filters, and a third block that makes the measured frequencies of the output waveforms of the frequency selection filters correspond to quantized frequency channels. a fourth block that detects the amplitude of the output waveform of each of the frequency selection filters; a channel register block that registers the amplitude of each output waveform of the fourth block to the frequency channel to which the waveform belongs; It is equipped with a self-channel register block that registers the center frequency response of each of the above-mentioned frequency selection filters corresponding to the channel, and a separately excited channel register block that registers the other responses, and is designed to analyze the frequency of the input signal. A waveform analyzer characterized in that it is configured as follows.