JPS61100800A - Signal analyzer - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an improvement in a signal analysis device used for frequency analysis of audio signals and other signals.

Vowels, which are the elements that make up speech, are sound waves generated by the vibration of the vocal cords sent out through the oral cavity, and are composed of several characteristic frequency bands, or formants, determined by the state of the oral cavity.

For this reason, for example, in speech recognition using a phoneme as a recognition unit, a process of obtaining a speech spectrum from a speech signal by frequency analysis is used in order to obtain formants.

In this case, it is important that one frequency analysis is obtained accurately and quickly in real time.

[Conventional technology]

FIGS. 2(a) and 2(b) are block diagrams showing the configurations of a first conventional example and a second conventional example of a signal analyzer, respectively, and hereinafter, the same reference numerals refer to the same objects throughout the plan. In FIG. 2(a), 1 is an AD converter that converts an audio signal from an analog signal to a digital signal, and 2 is an AD converter that converts an audio signal from an analog signal to a digital signal. Window processing section to cut out by width, 3 is window processing section 2
This is a frequency analysis section that analyzes the frequency of the signal extracted by, for example, processing such as fast Fourier transform, and obtains a frequency spectrum.

In FIG. 2 (bl), 4 is a least squares processing unit that approximates the signal cut out by the window processing unit 2 to a straight line by the least squares method, and 5 is a least squares processing unit from the signal cut out by the window processing unit 2. This is a subtraction unit that subtracts the signal obtained by 4.

[Problem that the invention seeks to solve]

According to the first conventional example, although the −Fourier transform is originally intended to perform arithmetic processing on the range from −■ to +■ of the signal to be transformed, Processing is performed only on signals with a window width of time length.

Therefore, if the original signal contains a component with a period longer than this window width, as illustrated in FIG. 2(C),
If the window processing section 2 cuts out the slope part (2) of the original signal (2), the resulting spectrum waveform (not shown) is nothing but that obtained from a periodic waveform like (2).

In other words, the frequency obtained by the frequency analysis section 3,
There is a problem in that a thousand spectra corresponding to a sawtooth wave like ■ appear superimposed on the vector.

In the second conventional example, in order to eliminate the problems in the first conventional example, a least squares processing unit 4 is installed before the frequency analysis unit 3.
and a subtraction section 5.

As explained with reference to FIG. 2(d), the portion of the signal S cut out by the window processing unit 2 is approximated as a straight line S° by the least squares method, and a straight line component s1 is extracted from the original signal S.
After converting it like sIF by subtracting 1
A frequency spectrum is obtained by the frequency analysis section 3.

However, in the case of voice recognition, for example, the AD converter 1
The sampling interval in is set to 0.1 ms.

Assuming that the window width used in the window processing section 2 is 30 milliseconds,
In the least squares processing section, about 900 multiplications and about 150
Since it is necessary to perform zero additions, there is a problem that the processing takes a long time and is therefore not suitable for real-time processing.

[Means for solving problems]

The signal analysis device according to the present invention includes a window processing unit that cuts out a signal with a window width of a predetermined time length, and performs the window processing using an average value of each portion of the first half and the second half of the signal cut out by the window processing unit. The above problem is solved by including a signal correction section that corrects the slope of the signal extracted by the signal correction section, and a frequency analysis section that frequency-analyzes the signal corrected by the signal correction section to obtain a frequency spectrum. It is something that

[Effect]

That is, as illustrated in FIG. 1(b), the window processing section 2
The signal of the part cut out by 5=(xi), where i=o ~(N+1)------
---'-(1), and the straight line yi = passing through the first half (the range from i=o to (N/2-1) and the second half (the range from i=N/2 to (N-1))] 2 (al-ao) (i-N/4
) /N a 0 is the slope of the signal (xi).

xi' = xi −yi −−−−−
-(5) is used to remove the slope of the signal (xi).

〔Example〕

The gist of the present invention will be specifically explained below with reference to an embodiment shown in FIG.

FIG. 1(a) is a block diagram showing the configuration of an embodiment of the present invention, and 61, 62, 63, and 64 represent respective parts in the first half and the second half of the signal cut out by the window processing section 2. This is a component of the signal correction unit 6 that corrects the slope of the signal cut out by the window processing unit 2 using the average value, and 61 is a first component that calculates the average value ao of the first half of (xi) using equation (2). The arithmetic unit of , (xi
) a second calculation unit for calculating the average value a1 of the latter half of 63
is (N/4) according to equation (4).

ao) and (3N/4, al), and 64 is a subtraction unit that performs subtraction according to equation (5).

Same as the second conventional example, if the sampling interval in the AD converter 1 is 0.1 milliseconds and the window width in the window processing section is 30 milliseconds, then 1 In the above embodiment, two divisions and 300 divisions are performed. Since the straight line of equation (4) is obtained by adding , the calculation time is significantly reduced compared to the second conventional example, and real-time processing becomes possible.

〔Effect of the invention〕

As explained above, according to the present invention, the slope of the extracted signal can be removed by window processing, and since the time required for the processing is short, real-time processing is possible.

[Brief explanation of drawings]

FIG. 1(a) is a block diagram of an embodiment of the present invention. Fig. 1 (bl is an explanatory diagram of the operation of the same embodiment. Fig. 2 (al is a block diagram of the first conventional example.) Fig. 2 (bl is a block diagram of the second conventional example. C) is an explanatory diagram of the first conventional example. Fig. 2 (d) is an explanatory diagram of the second conventional example. In the figure. 1 is the to conversion section, 2 is the window processing combination. (c) Ne Z leg (dn

Claims (1)

[Claims] A window processing section that cuts out the signal with a window width of a predetermined time length, and a slope of the signal cut out by the window processing section based on the average value of each part in the first half and the second half of the signal cut out by the window processing section. A signal analysis device comprising: a signal correction section that corrects the signal; and a frequency analysis section that frequency-analyzes the signal corrected by the signal correction section to obtain a frequency spectrum.