JP2672946B2 - Vocal training machine - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to an apparatus for practicing vowel vocalization by a speech-impaired person with speech disabilities. [Prior Art] In a conventional vocal training machine, a first formant and a second formant are extracted from an audio signal detected by a detection unit such as a microphone by an extraction unit, and a first formant and a second formant are extracted by a determination unit. By judging whether or not it is within the proper range, it is judged whether or not the uttered sound uttered by the trainee is within the target uttered sound range. Then, the judgment result was displayed to the trainee via a display means such as a display. [Problems to be Solved by the Invention] However, in the conventional technique, as shown in the graph of FIG. 4 in which the horizontal axis represents the first formant and the vertical axis represents the second formant, The voicing sound of "" exists near the same first formant, and the second formant of "a" almost coincides with the overtone of the second formant of "u". Similarly, the vocal sounds of "e" and "o" also exist in the vicinity of the first formant, and the second formant of "e" almost coincides with the overtone of the second formant of "o". On the other hand, depending on the speaker, a peak may occur between the first formant and the second formant in the uttered sound, or the peak of the second formant may be low. Therefore, when the trainee trains the vocal sound of, for example, "a", the conventional vocal training machine erroneously determines the peak uttered between the original first formant and the second formant as the second formant. , If the trainee trains the vocal sound of “U”, on the contrary, if the peak of the second formant is low, the second formant is not judged as the second formant, and 3 formants are erroneously determined to be the second formants, resulting in erroneous determination of "a", such as "a" and "u" or "d" and "o".
It has a problem that it is easy to make an erroneous determination. The present invention has been made in view of the above circumstances, and an object thereof is to provide a vocal training machine that can detect a vocal sound of a vowel without making an erroneous determination. [Means for Solving Problems] The detection means (20) for detecting the voiced sound of the trainee as an audio signal and the audio signal detected by the detection means (20)
Formant extraction means (51) for extracting the formants and the second formants, and the first mode for performing the formant extraction means (51) for the vocal training of "a", "e", "a", and "a", "a" The mode selector switch (38) capable of switching to the second mode for performing the "O" and "U" vocal training, and the Japanese vowel "A" from the first and second formants extracted by the formant extraction means (51). ",
A discriminating means (53) for discriminating among "a", "u", "e", and "o", and a display means (40) visually showing the discrimination result of the discriminating means (53). The formant extraction means (51) obtains the bandwidths of the third peak and the second peak from the low frequency side on the frequency axis of the audio signal, and extracts the first peak as the first formant. If the bandwidth of the second peak is smaller than the discrimination reference bandwidth, the second peak is extracted as the second formant, and if it is larger, the third peak is extracted as the third formant, and the first mode and the second mode are extracted. The technical means is to change the discrimination reference bandwidth. [Operation and Effect of the Invention] According to the present invention, the extraction means are “a”, “d”, and “a”.
The second formant is extracted by switching between the first mode and the second mode of "a", "o", and "u". Specifically, the bandwidth for discriminating the second formant is changed depending on the set first mode and second mode, and if the bandwidth of the second peak lower than the lower frequency is smaller than the bandwidth of the discrimination reference, it becomes 2 The third peak is discriminated as the second formant, and when the bandwidth of the second peak is not smaller than the discrimination reference bandwidth, the third peak is discriminated as the second formant from the lower frequency. In this way, by determining the second formant by the extraction means, the second formant has a harmonic relationship.
Since "U" and "A" and "E" and "O" are detected in different modes, the second formant of each vowel can be reliably identified. Therefore, it is possible to surely prevent erroneous determination of “U” and “A” and “D” and “O”. [Embodiment] Next, a vocal training machine of the present invention will be described based on an embodiment shown in the drawings. FIG. 2 shows a block diagram of the hardware of the vocal training machine. The voicing training machine 10 is a microphone 20 which is a detection means of the present invention for detecting a voicing sound of a trainee as a voice signal, a signal processing unit 30 for electrically processing the signal of the microphone 20, and a signal processing unit 30 The color monitor television 40 is the display means of the present invention for displaying the output information to the trainee. The signal processing unit 30 is a low-pass filter 31 that cuts a high frequency among frequency components of a sound signal from a sound signal detected by the microphone 20, and a high-frequency cut analog audio signal output from the low-pass filter 31 as a digital signal. An analog / digital conversion unit (hereinafter A /
D converter) 32 and random access memory (hereinafter RA
M) 33, central processing unit (hereinafter CPU) 35 equipped with a read-only memory (hereinafter ROM) 34 of a cartridge type that can be replaced according to the age, gender of the user, purpose of use, etc.
And digital data processing that transfers data to RAM33 and CPU35, and performs formant extraction from the voice signal produced by the trainee from the digital signal input to CPU35 (hereinafter referred to as DS
P) 36 and a graphic processing 37 for displaying the calculation result of the CPU 35 on the color monitor television 40. The vocal training machine 10 uses the Japanese vowels "A", "I",
"U", "E", "O" vocal training,
"A", "E", "A" vocalization practice and "A",
The training is divided into two series of "O" and "U" utterance practice. The CPU35 has the first mode of "A", "E", "A" and "A", "O". , And the second mode of "C" is provided. As shown in the software block diagram of FIG. 1, the microcomputer unit 50 including the RAM 33, the ROM 34, the CPU 35, and the DSP 36 analyzes the signal of the A / D conversion unit 32 and outputs a sound within a predetermined width from the low frequency. A formant extraction section 51 which is an extracting means for detecting a pressure peak and extracting a formant according to the mode changeover switch 38, and stores the data of the first formant and the second formant of each vowel, and CP
It comprises a storage unit 52 for storing the control order of U35 and the like, and a discriminating unit 53 which is a discriminating means of the present invention for discriminating a voicing sound from the first formant and the second formant extracted by the extracting unit. . Next, the formant extraction processing in the formant extraction unit 51 will be described with reference to the flowchart of FIG. When the digitized audio signal is input from the A / D conversion unit 32, in step S1, the digital signal is analyzed and three peaks P1, P2, P3 and the bandwidth of the peaks are analyzed from the low frequency side. Find B1, B2, and B3 (see Figure 4). Next, in step S2, the mode changeover switch 38
Determines whether the first mode is set. If the result of the determination in step S2 is YES, in step S3, the bandwidth BJ that is the criterion for the formant extraction is determined.
Is set as a constant C1. In addition, the determination result of step S2 is NO.
In this case, in step S4, the bandwidth BJ that is the criterion for formant extraction is set as a constant C2. The bandwidth of the constant C2 is set larger than the bandwidth of the constant C1. Next, in step S5, the frequency of the peak P1 obtained in step S1 is recognized as the first formant F1. Next, in step S6, it is determined whether or not the second bandwidth B2 from the lower frequency side is smaller than the bandwidth BJ set in step S2 or step S3. If the decision result in the step S6 is YES, in a step S7, the frequency of the second peak P2 from the low frequency side obtained in the step S1 is recognized as the second formant F2, and then the vowel determination of the discrimination section 53 is performed. move on. If the determination result in step S6 is NO, in step S8, step S1
The frequency of the third peak P3 from the low frequency side obtained in step S3 is recognized as the second formant F2, and then the vowel determination of the determination unit 53 is performed. Next, the vowel determination process of the determination unit 53 will be described based on the flowchart of FIG. The discriminating unit 53 of the present embodiment is performed by a tree determination method using a linear discriminant function based on Mahalanobis general distance. First, in step S9, the first formant and the second formant extracted by the formant extracting unit 51 are extracted. The formant determines whether the group belongs to "a" or "a". If it is determined in step S9 that "A" is present, in step S10 it is determined which of the groups "A" and "U" the first formant and the second formant extracted by the formant extraction unit 51 belong to. Make a decision. When it is determined to be “A” in step S10, the first formant and the second formant extracted by the formant extraction unit 51 in step S11 are
It is determined which of "a" and "d" the group belongs to. If it is judged as “A” in step S11,
In step S12, the first formant and the second formant extracted by the formant extraction unit 51 are
It is determined which group, "A" or "O", belongs to. If it is judged as “A” in step S12,
In step S13, it is determined that the utterer uttered by the trainee is "A". If it is judged as "a" in step S9, the step
In S14, the first extracted by the formant extraction unit 51
The formant and the second formant determine which group "a" or "u" belongs to.
If "a" is determined in step S14, step S
At 15, the first formant and the second formant extracted by the formant extraction unit 51 are “a” and “d”.
It is determined which of the groups to belong to. If "i" is determined in step S15, which of the groups "i" and "o" the first formant and the second formant extracted by the formant extraction unit 51 belong to in step S16. Make a decision. Step S16
If it is determined to be "a", it is determined in step S17 that the sounder uttered by the trainee is "a". If it is determined to be “U” in step S10 and step S14, in step S18, the formant extraction unit 5
It is determined whether the first formant and the second formant extracted in 1 belong to the group “U” or “D”. If it is determined in step S18 that "u" is present, it is determined in step S19 which of the groups "u" and "o" the first formant and the second formant extracted by the formant extraction unit 51 belong to. Make a decision. When it is determined to be “U” in step S19, it is determined in step S20 that the utterance sound of the trainee is “U”. If “d” is determined in step S11, step S15, or step S18, whether the first formant and the second formant extracted by the formant extraction unit 51 are “d” or “o” in step S21. Determine if it belongs to a group. If "d" is determined in step S21, it is determined in step S22 that the utterance sound of the trainee is "d". Step S12, Step S16, Step S19, Step S
When it is determined to be "o" in 21, it is determined in step S23 that the utterance sound uttered by the trainee is "o". Then, the determination result determined by the determination unit 53 is processed by the graphic processing 37 and displayed on the color monitor television 40. It should be noted that when the mode changeover switch 38 is set to the first mode, when it is determined to be "u" or "o", or when the mode changeover switch is set to the second mode, "a" is set. Alternatively, when it is determined to be “d”, the fact that the determination cannot be made is displayed on the color monitor television 40. In this embodiment, the band width of the discrimination criterion of the second formant is changed according to the setting of the mode changeover switch 38, so that the peak between the first formant and the second formant is erroneously changed to the first peak as in the conventional case. There is no possibility of discriminating between the two formants and erroneously discriminating the third formant from the second formant. As a result, it is possible to reliably prevent the erroneous determination of “a” and “u” and “d” and “o” as in the conventional case.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram of software of a vocal training machine, and FIG.
The figure is a block diagram of the hardware of the vocal training machine, FIG. 3 is the flowchart showing the operation of the formant extraction unit, and FIG.
FIG. 6 is a graph for explaining the embodiment in which the horizontal axis represents frequency and the vertical axis represents sound pressure power. FIG. 5 is a flowchart showing the operation of the discriminating unit. FIG. It is a graph showing the second formant and showing the distribution of each vowel. In the figure, 20 ... Microphone (detection means), 30 ... Signal processing section, 38 ... Mode changeover switch, 40 ... Color monitor TV (display means), 50 ... Microcomputer section, 51 ... Formant extraction section (Extracting means), 53 ... Discriminating unit (discriminating means)

Claims (1)

(57) [Claims] Detecting means (20) for detecting the voiced sound of the trainee as a voice signal, formant extracting means (51) for extracting the first formant and the second formant from the voice signal detected by the detecting means (20), and formant extraction The means (51) is "a", "d", "a".
The first mode to perform the vocal training of "A", "O",
A mode changeover switch (38) capable of switching to a second mode for carrying out a vocal training of "U", and a Japanese vowel "A" from the first and second formants extracted by the formant extraction means (51),
A discriminating means (53) for discriminating among “a”, “u”, “e”, and “o”, and a display means (4) visually showing the discrimination result of the discriminating means (53).
0) and the formant extraction means (51) has a third peak from the low frequency side on the frequency axis of the audio signal, and 2
Obtain the bandwidth of the second peak, extract the first peak as the first formant, and if the bandwidth of the second peak is smaller than the discrimination reference bandwidth, the second peak is the second formant, and if it is larger, A vocal training machine that extracts the third peak as a third formant and changes the discrimination reference bandwidth between the first mode and the second mode.