JPH0298000A - Method for identifying male and female voices - Google Patents

Abstract

PURPOSE:To easily identify male and female voices by calculating the referring axis crossing number of an input signal and a value related to the amplitude distribution of a waveform as a feature parameter and comparing the calculating result with the dictionary data of the male and female vocalized voices. CONSTITUTION:The dictionary data to make the referring axis crossing number between 20mS and the value related to the amplitude distribution of the waveform into the feature parameter for the male and female representative vocalized voices are stored into a storing part 16. For a signal gathered by a microphone 11, 4.2kHz or above is cut through an amplifier 12 by a low-pass filter 13, the signal is made into a digital signal with a sampling frequency 10kHz and a conversion bit number 16 bits by an A/D converter 14, and it is sent into a parameter calculating part 15. The referring axis crossing number and the value related to the amplitude distribution of the waveform are calculated as the feature parameter. Next, the feature parameter and the standard pattern of the dictionary data are compared by a deciding part 17, and the male and female voices are identified.

Description

[Detailed description of the invention] [Industrial application field] The present invention relates to a method for identifying male and female voices.

[Prior Art] Conventionally, as a method for identifying male and female voices,
7600, which uses the difference in pitch frequency between men and women, or JP-A-60-1 [12298]
There is a method using the difference between men and women in the first and second formant frequencies, as reprinted in the above publication.

[Invention or problem to be solved U] However, none of the above methods can be said to have high discrimination ability because they are complicated and take a long calculation time, and the overlap between male and female voices on the frequency axis is large. There are difficulties in developing it for general use.

In addition, in the identification method based on formant frequency, calculation time can be shortened by detecting the formant frequency using dedicated hardware (multiple band-pass filters), but the non-formant poles are It may be falsely detected as a pole, and high discrimination ability cannot be expected.

An object of the present invention is to achieve high discrimination ability using a simple method when identifying male and female voices.

[Means for Solving Problem 5B] The present invention according to claim 1 calculates the number of reference axis crossings of the input signal and the value related to the amplitude distribution of the waveform as characteristic parameters, and uses this calculation result as a Fl
It was designed so that

According to a second aspect of the present invention, a wave height value P expressed by the following equation is used as a value related to the amplitude distribution of the waveform.

P = 20x Iog+o(Vp/Vr+w*), where V p is the maximum absolute value of the amplitude within a certain time interval V 1BB + the effective value of the amplitude within the same certain time interval. In this example, a wave height value P expressed by the following equation is used as a value related to the amplitude distribution of the waveform.

P = 20x log+o(Vp/Va), v
P: maximum value of the absolute value of the amplitude within a certain time interval v4: average value of the absolute value of the amplitude within the same certain time interval This method uses the time (referred to as super-standard amplitude time) during which the effective value exceeds a threshold value within a time interval.

[Operations] In the present invention as set forth in claim 1, male and female voices are identified as follows. In addition, in the present invention, voiced sounds (vowels,
There are sounds that involve vibration of the vocal cords, such as semi-vowels and nasal sounds, and almost all sounds produced by humans include voiced sounds and are considered to be speech.

(1) For male and female voices, the number of reference axis crossings (such as zero level, etc.) within a certain time interval and the value related to the waveform amplitude distribution are defined as characteristic parameters. Prepare the word M data to be used.

Note that the above dictionary data includes numerical data obtained by converting acoustic data into feature parameters, average values obtained by statistical processing of numerical data,
The value can be determined in various ways, such as statistical data such as dispersion, or discriminant data such as a boundary equation determined based on statistical data.

(2) An input signal is sampled, and the number of reference axis crossings within a certain time interval of this input signal and the value related to the amplitude distribution of the waveform are calculated as characteristic parameters.

(3) The feature parameters calculated in (2) above and the above (
The standard pattern defined by the dictionary data defined in 11.
Comparisons are made in the parameter space, and pattern recognition is used to determine whether the input signal is a male or female voice.

In other words, the category "male voice" defined by dictionary data
and the category "female voice", the parameter space is divided into two, and by using a -f-like pattern recognition method, it is determined whether a male voice is a female voice by determining which category the characteristic parameters of the human voice fall into. . Note that the dictionary data is a set of a sufficient number of feature parameters for representative voiced sounds.

However, in the present invention recited in claim 1, two parameters, the number of reference axis crossings and the value related to the amplitude distribution of the waveform, are used as feature parameters, so the category "male voice" and the category "female voice" are separated. Can be clearly separated in parameter space.

Therefore, male voices and female voices can be easily identified with a high identification rate.

According to the present invention as set forth in claim 2, since the wave height value as described above is used as the value related to the amplitude distribution of the waveform, a parameter value that exhibits a remarkable difference between men and women in the sharpness of the waveform is used, This has the advantage of improving the discrimination between male and female voices.

According to the present invention as set forth in claim 3, since the above-mentioned wave height value is used as the value related to the amplitude distribution of the waveform, the amount of calculation is reduced compared to the present invention as set forth in claim 2, and the waveform This method uses parameter values that exhibit a noticeable difference between men and women in terms of sharpness, which has the advantage of improving the distinguishability of male and female voices. According to the present invention as set forth in claim 4, in which a small amount of calculation means a fast response speed, the super-standard amplitude time as described above is used as the value related to the amplitude distribution of the waveform. Compared to the present invention described in item 2 or 3, there is an advantage 1- that the amount of calculation can be further reduced.

[Embodiment] FIG. 1 is a block diagram showing an example of a gender discrimination device used for implementing the present invention, and FIG. 2 is a schematic diagram showing a parameter space formed by characteristic parameters of the present invention.

In FIG. 1, 11 is a microphone, 12 is an amplifier, 13 is a low-pass filter, 14 is an A/D converter, 15 is a parameter calculation section, 16 is a dictionary data storage section, 17 is a judgment section, and 18 is a result output section. . In this example,
Male and female voices are detected as follows.

(1) For the representative voiced sound [a] of men and women, prepare dictionary data whose feature parameters are the number of reference axis crossings in 20 m5 of these signals x 1 and the value regarding the amplitude distribution of the waveform x 2, This is stored in the dictionary data storage section 16.

Here, as the value x 2 regarding the amplitude distribution of the waveform, any one of the following ■, ■, and ■ can be used.

■The wave height value P expressed by the following formula.

P = 20x log + o (Vp/vt, ssl plus vpcony - Maximum absolute value of amplitude within the time interval vr □ ; Effective value of amplitude within the same fixed time interval■ Peak value P expressed by the following formula .

p = 20 x log + o (Vp/v, also vPcony - Maximum absolute value of amplitude within a time interval v6: Average value of absolute value of amplitude within the same fixed time interval ■ Amplitude or within a fixed time interval The time required to exceed the threshold value for determining the effective value (super amplitude time).

When using the wave height value of (2) above, a parameter value that exhibits a significant difference between men and women in the sharpness of the waveform is used, which has the advantage of improving the discriminability of male and female voices.

When using the wave height value in (■) above, the amount of calculation can be reduced compared to the wave height value in (■) above, and a parameter value that exhibits a noticeable difference between men and women in the sharpness of the waveform is used. This has the advantage of improving identifiability.

When using the wave height value of (2) above, there is an advantage that the amount of calculation can be reduced compared to the wave height values (2) and (4) above.

(2) Collect the human input signal F) with the microphone 11, amplify this human input signal with the amplifier 12, pass it through the low-pass filter 13 to cut off components of °r4.2KIIz or more, and send it to the A/D converter 14. The sampling frequency is 10KI by
Iz is converted into a digital signal with a conversion bit count of 16 bits and sent to the parameter calculation section 15.

The parameter calculation unit 15 calculates Z O+n of the human input signal.
Sl! The number of reference axis crossings in fl×1 and the value regarding the amplitude distribution of the waveform×2 are calculated as feature parameters.

(3) The feature parameters calculated in (2) above and the above (
11, the dictionary data or the standard pattern specified in
The determination unit 17 compares and determines whether the human input signal is a male voice or a female voice, and outputs the determination result from the result output unit 18.

Here, pattern recognition is performed using the aforementioned dictionary data.

The identification is made, for example, on the parameter space shown in FIG. 2 as follows.

That is, in FIG. 2, two characteristic parameters, the number of zero crossings (the reference axis level is set to zero level) and the peak value, are plotted on the Xl axis and the x2 axis, respectively. In Figure 2, μlx I++ σ12 represents the average value of the male voice dictionary parameters, the standard deviation of the xtN component, and the standard deviation of the ×22 component, and μ2, σ,
1 and σ22 represent similar values for female voice dictionary parameters, respectively.

A broken line A represents the concept of the category "male voice" defined by μ and σ, and a broken line B represents the concept of the same category "female voice."

In addition, the boundary C defines the feature parameter space with the category “So(
This is the boundary dividing the category "female voice J" into the category "female voice J," and the side containing the average value μm of the Mn data of the male voice is the category "male voice." Boundary C is between the category “male voice” and the category “
It is a collection of points that have the same likelihood for "female voice". In the case of this embodiment, the category "male voice" is a closed space because the standard deviation of the male voice dictionary data is smaller than the standard deviation of the female voice dictionary data. When the feature parameter calculated from the input voice taken into the calculation unit 15 belongs to the μm side of the above boundary C on the feature parameter space, it is determined that the input voice is a male voice.

However, in the above embodiment, since two parameters, the number of reference axis crossings and the value related to the amplitude distribution of the waveform, are used as feature parameters, the category "male voice" and the category "female voice" can be clearly separated in the parameter space. can. Therefore, male voices and female voices can be easily identified with a high identification rate.

Furthermore, the above feature parameters can be calculated in a short calculation time without using special hardware, and can be easily applied to general applications.

In the above embodiment, a collection of points with equal likelihoods for two categories was used as the boundary line defining the standard pattern on the feature parameter space, but of course other general methods may be used in implementing the present invention. For example, if the category ``
Instead of a collection of points that have the same likelihood for the category "male voice" and the category "female voice," a collection of points that have the same Maharanobis distance or Euclid distance can be used.

[Effects of the Invention] As described in [2] above, according to the present invention, it is possible to identify male and female voices in a short calculation time and without using special hardware.
! There are two ways to achieve high discrimination ability.

[Brief explanation of drawings]

Figure 1 shows a male and female voice identification device used to implement the present invention: rt
FIG. 2 is a block diagram showing an example of the above, and FIG. 2 is a schematic diagram showing a parameter space formed by the characteristic parameters of the present invention. DESCRIPTION OF SYMBOLS 11... Microphone, 15... Parameter calculation part, 16... M data storage part, 17... ↑-reference part, 18... Result output part. Figure 1 Figure 2 Patent applicant: Sekisui Chemical Co., Ltd. Representative Hiroshi 1) Kaoru

Claims (4)

[Claims] (1) Calculate the number of reference axis crossings of the input signal and the value related to the amplitude distribution of the waveform as characteristic parameters, and compare the calculation results with dictionary data of male and female voiced sounds to determine whether the input signal is uttered by either a male or a female voice. A method for identifying male and female voices. (2) The male and female voices according to claim 1, wherein the value regarding the amplitude distribution of the waveform is a peak value expressed as a ratio of the effective value of the amplitude within the certain time interval to the maximum value of the absolute value of the amplitude within the certain time interval. How to identify. (3) The wave height value expressed as the ratio of the average value of the absolute value of the amplitude within the certain time interval to the maximum value of the absolute value of the amplitude within the certain time interval is used as the value related to the amplitude distribution of the waveform. How to identify male and female voices. (4) The method for identifying male and female voices according to claim 1, wherein the value related to the amplitude distribution of the waveform is a time period during which the amplitude exceeds a threshold value based on an effective value within a certain time interval.