JPH0812535B2 - Pronunciation training device - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a pronunciation training device that effectively applies pronunciation training for English and the like by applying a voice analysis technique.

(Prior Art) Conventionally, when training pronunciation of a foreign language, there is a method of relying on a person whose native language is a foreign language for pronunciation evaluation, and a method of self-training with a commercially available cassette tape, video tape, record, or the like. There is.

(Problems to be Solved by the Invention) The above method of relying on a person whose native language is a foreign language for pronunciation evaluation has a disadvantage that it is difficult for that person to explain a subtle difference in pronunciation to a learner. . None of the methods using a cassette tape, a video tape, etc. has a method for confirming whether or not the pronunciation is close to the correct pronunciation, and there is an inconvenience that the learner's hearing must judge and evaluate his own pronunciation.

Therefore, the present applicant first analyzed the voice associated with his or her own pronunciation, and displayed on the display the voice parameters such as the waveform, power, pitch (pitch of sound), sound spectrograph, etc., and the voice Compare the pronunciation feature data measured from the parameter pattern with the model pronunciation feature data, obtain a corrective comment corresponding to the result from the comparison result, and stress accent to efficiently approach the standard voice pattern by the teacher, intonation,
We proposed a pronunciation training device for training pronunciation of vowels and consonants (see Japanese Patent Application No. 61-303772). The present invention relates to an improvement of the pronunciation training device, and an object thereof is to provide a pronunciation training device capable of more efficiently performing pronunciation training.

(Means for Solving the Problems) In order to achieve the above object, the pronunciation training apparatus of the present invention has a voice signal generating means and a voice for analyzing a voice associated with pronunciation from a voice signal as described in claim 1. An analysis means, a memory for storing at least voice data and a voice analysis result, a display having a pair of screens of an effective screen in an upper half and a lower half, and at least the same training item at the same relative position corresponding to each screen. Pronunciation training provided with two sets of key groups consisting of a plurality of operation keys arranged in a row, and means for comparing and displaying voice analysis results by independently controlling a pair of screen configurations on the display by operating the key groups. In the device, the pitch value data written in the memory is sequentially read out, and when the pitch value data is a voiced sound, the pitch value data is displayed on the display. , If the read pitch value data is voiced and the next read pitch value data is silent or unvoiced,
Intonation analysis operation means for reproducing the voice corresponding to the pitch value data of the voiced sound read continuously until then and performing the series of operations again for the remaining pitch value data after a lapse of a predetermined time is provided. As described in claim 2, in the pronunciation training apparatus according to claim 1, the memory has a voice memory for storing voice data and a voice rising memory, and a write operation is circulated to the voice rising memory. Then, the rising portion of the voice data is written in the voice rising memory, and when the voice data exceeds a certain value, the writing operation is switched to the voice memory, and the unstored area of the voice memory corresponding to the voice rising memory is continued. After writing the voice data, the voice data written in the voice rising memory after the writing is finished is written in the voice memory. 3. The sounding training device according to claim 1, further comprising a recording operation means for transferring to a non-stored area, wherein the signal transmission circuit of the sound signal generating means other than the microphone is provided with a recording level adjusting volume. And

(Operation) When the teacher and the learner operate the teacher key and the learner key, respectively, a pair of screens showing the voice analysis result displayed on the display are controlled independently of each other.

When performing intonation analysis or sound spectrograph analysis, voice data is read from the voice memory stored in the voice memory, passed through a filter, converted into voice data suitable for the analysis, and written into the working memory. The voice analysis is performed on the voice data stored in the working memory.

When recording the audio data output from the audio signal generating means, the rising portion of the audio data is written in the audio rising memory which is circulated through the writing operation, and is written in the audio memory when the audio data exceeds a certain value. .
After the writing of the voice data to the voice memory is completed, the rising portion stored in the voice rising memory is written in an unstored area of the voice memory corresponding to the voice rising memory.

When training intonation, the pitch value data stored in the memory for the long sentences to be pronounced in a fluent manner are sequentially read, and when the read pitch value data is voiced, the pitch value data is displayed on the display. When the displayed and read pitch value data is voiced and the next read pitch value data becomes unvoiced or unvoiced, the pitch value data of the voiced sound read continuously until then The voice corresponding to is reproduced, and after a lapse of a predetermined time, the series of operations is performed again for the remaining pitch value data.

The output of the microphone of the audio signal generating means is stored in the audio memory without its size being adjusted.

Embodiment An embodiment of the present invention will be described below with reference to the drawings.

 FIG. 1 shows the appearance of the pronunciation training apparatus of the present invention.

In the figure, 1 is a personal computer (hereinafter referred to as a personal computer), 2 is a display, 3 is a dedicated keyboard, 4 is a unit accommodating an external circuit (hereinafter referred to as a dedicated unit), 5 is a microphone, 6 is a speaker, and 7 is a hard copy. It is a well-known printer that obtains a print output.

As shown in FIG. 2, the special keyboard 3 has a normal alphanumeric key 10 in the center, and a pair of special keys for the teacher (hereinafter referred to as the SCR1 key) 11 and for the learner. A key (hereinafter referred to as SCR2) 12 is arranged and configured. Both the SCR1 key 11 and the SCR2 key 12 are composed of a plurality of operation keys 11-1 to 11-17 and 12-3 to 12-17. The difference between them is that the SCR1 key 11 and the BEGIN key 11-1 are different. And END
With the key 11-2, the other keys are arranged in the same way. The BEGIN key 11-1 is a key for inputting the start of training, the END key 11-2 is a key for inputting the end, and these keys are provided only on the SCR1 key because the operation is judged by the teacher. To leave it to.

Since the total number of keys is not much different from that of general computer keyboards, the positions and names of keys other than the alphanumeric keys arranged in the center of the computer keyboard are shown in SC in the second illustration.
It only needs to be changed like the R1 key 11 and SCR2 key 12.
Therefore, the key scanning of the dedicated keyboard 3 and the key information communication with the personal computer 1 are performed by a known method.

As shown in FIG. 3, the personal computer 1 has a CPU 13, a memory 14, and a disk 15 as main constituent elements.

The memory 14 is used as a processing program and buffer memory, a control parameter / flag memory, a voice rising memory, a voice memory, a working memory and an image memory. The processing program and the buffer memory are SCR1 keys.
11 and SCR2 key 12 are provided in common, other memory is
The SCR1 key 11 and the SCR2 key 12 are provided separately.

The voice memory stores the digital sample value of the voice signal to be analyzed, and the voice rising memory temporarily stores the digital sample value of the voice signal of the voice rising.
The digital sample value is transferred to the corresponding area of the voice memory upon completion of recording. The working memory stores a digital value of a voice signal in which a high frequency component is emphasized in order to express the result in a well-balanced and uniform shade when obtaining a sound spectrograph analysis (hereinafter referred to as a pattern analysis), or Stores a voice signal with unnecessary high frequencies reduced and sample intervals reduced so that the fundamental frequency component, which is the pitch, can be easily extracted when a pitch analysis (hereinafter referred to as intonation analysis) of the voice is obtained. is there. The image memory stores the contents displayed on the display 2.

The processing of the personal computer 1 is performed by giving priority to the SCR1 key 11 or the SCR2 key 12 that is pressed first.

A series of processes for performing voice analysis from the recorded digital sample value of the voice memory, storing the result in the image memory and displaying it on the display 2 are described in the above-mentioned Japanese Patent Application No. 61-3037.
Since it is described in the specification of No. 72, it will not be described in detail.

The detailed appearance of the dedicated unit 4 is shown in FIGS. 4A and 4B, and the external circuit housed therein is shown in FIG.

In FIGS. 4 and 5, reference numeral 16 is an input selector, and the input selector 16 is composed of switches A1, A2, and M, and the switch A1 is an input used when recording the audio output of a tape recorder (not shown). Terminal AUX1 and external circuit, switch
A2 is an input terminal AUX2 used for recording the audio output of a video disc (not shown), an external circuit, and a switch M.
Is for selectively connecting the input terminal MIC used for recording the voice output of the microphone 5 and the external circuit. When the switch A1 is turned on, the line L17 becomes the high level "H" and the electronic switch 17 Turns on and the input terminal
The input signal of AUX1 is input to the buffer amplifier 18. When the switch A2 is turned on, the line L19 becomes "H", the electronic switch 19 is turned on, and the input signal of the input terminal AUX2 is input to the buffer amplifier 18. When switch M is turned on,
The line L20 becomes “H”, the electronic switch 20 is turned on, and the input signal of the input terminal MIC is input to the buffer amplifier 22 via the buffer amplifier 21. When the switch M is off, the line L20 is "L", the electronic switch 20 is off, and the output of the NOT circuit 23 is "H", so the electronic switch 24 is on. Therefore, when the switch A1 or A2 is turned on, the input signal of the input terminal AUX1 or AUX2 is input to the buffer amplifier 22 via the buffer amplifier 18.

When the input signals from the input terminals AUX1 and AUX2 are input to the buffer amplifier 22, the recording level adjustment volume 25 (LEVEL VO
L.) adjusts the size of the audio signal, whereas when the input signal of the input terminal MIC is input to the buffer amplifier 22, the input signal does not pass through the volume 25. ing. Therefore, it is possible to determine the loudness of the voice to be pronounced based on the loudness of the input signal. The loudness of voice is important for pronunciation training, and one of the problems is the lack of voice volume when Japanese speak English. In other words, Japanese with chest breathing is inevitably inadequate in power of English caused by abdominal breathing. Therefore, when the waveform of the sound produced when the sound is produced with the distance between the mouth and the microphone 5 being fixed is seen on the display 2, the amplitude is dotted lines LV and LV as shown in (α) of FIG.
It is possible to judge that the voice volume is insufficient when the interval is entered, and when the amplitude is sufficiently out of the dotted lines LV and LV as shown by (β), it is determined to be a pass.

Incidentally, although the microphone 5 may be a desk-top type as shown in the first illustration, there is no problem, but if it is a method of putting on the head integrated with the headphone, it is convenient because the distance relationship between the mouth and the microphone 5 can be fixed.

The output of the buffer amplifier 18 is the buffer amplifier.
22 and a monitor signal to a summing amplifier 26, the output of which is a low-pass filter 27 for removing sampling noise and a speaker volume 28 (VOLUME).
Power amplifier 2 while sending to the output terminal REC.OUT via
Input to 9 and its output is connected to speaker 6. Output terminal S
P.OUT and a pair of headphones volume 30 ₁ , 30 ₂ (PH.1 V
OL. And PH.2 VOL), and outputs to the output terminals PHONE1, PHONE2 that connect headphones. The pair of output terminals PHONE1 and PHONE2 are SCR1 key 11 and SCR2 key 12
Is prepared for each user, and the volume can be set to his or her own preference by providing the headphone volumes 30 ₁ and 30 ₂ . The low-pass filter 27 has a cutoff frequency of 6 KHz and a characteristic of about -24 dB / OCT, for example.

31 is used to connect the external circuit to the PC 1
An I / O interface connector, to which a decoder 32 (DEC) is connected. The decoder 32 is a port designation signal on the address line input from the personal computer 1.
Read pulse to decode input command pulse IOR and output command pulse IOW to supply to input port 33 (IN1) and write to supply to output port 34 (OUT1), output port 35 (OUT2) and output port 36 (OUT3) respectively It generates a pulse. In the input line of the input port 33,
The A / D converter 37 (ADC) is connected to the A / D converter 37.
Is a summing amplifier 38 and an electronic switch 40 through an anti-aliasing filter 39 having a characteristic of about 6 KHz, −70 db, for example, which cuts off frequencies beyond 6 KHz, which is the frequency band of the voice sound. It is connected to 41 and converts the analog audio signal input from the addition amplifier 38 into a digital audio signal in accordance with the output signal of the 12 KHz main oscillator 42. Output port of output port 34 Q ₁ , Q
_2, Q _3, Q ₄ in the ...... Q _P, the light emitting diode of the recording time indicator 60 for displaying the recording time (REC.TIME) (LED) 6
1 ₁ , 61 ₂ , 61 ₃ ... are connected, and an AND circuit 62 and an OR circuit 63 are interposed between the one light emitting diode 61 ₁ and the output terminal Q _1, and the-input terminal of the AND circuit 62 is connected. The main oscillator 42
It is connected to the frequency divider 64 that makes the output frequency of 1/4096 to 3Hz. D- via the latch 65 to the output port 35
An A converter 66 (DAC) is connected, and the D-A converter 66 is connected to the summing amplifier 26 via an electronic switch 67, and emphasizes a high frequency component of voice necessary for pattern analysis. Is connected to the electronic switch 41 via a high-pass filter 68, and via a low-pass filter 69 and an electronic switch 70 for reducing unnecessary high frequency components other than the fundamental frequency component which is the pitch required for intonation analysis. Connected to the summing amplifier 38. Output port REC of output port 36 (OUT3) is "H" during recording operation
Then, the electronic switch 40 is turned on to input the audio signal into the AD converter 37, and the electronic switch 71
Is turned on to drive the level indicator 72 (LDS), and a plurality of LEDs 74 constituting the level meter 73 are turned on according to the input level. The output terminal PI of the output port 36 becomes "H" at the time of intonation analysis operation, whereby the electronic switch 70 is turned on and the audio signal output from the output port 35 is passed through the low pass filter 69 to the AD converter 37.
(ADC), the output terminal SG of the output port 36 becomes “H” during the pattern analysis operation, thereby turning on the electronic switch 41 and transmitting the audio signal output from the output port 35 through the high pass filter 68. It is adapted to be input to the A / D converter 37.

In FIG. 4, 75 is a power cord and 76 is a power switch.

As shown in FIG. 6, the SC is displayed on the display 2.
A pair of screen configurations SCREEN1, SCREEN of the upper half and the lower half which are independently controlled by the R1 key 11 and the SCR2 key 12, respectively.
2 is displayed. The coordinates of the main points are entered on the display 2 shown in FIG.
Since it is sufficient to use a personal computer 1 or a display 2 having a graphic capability of 1024 points and a Y axis of about 348 × 2 points, a general commercial product will suffice, and an intermediate class personal computer will be good and low cost.

As shown in FIG. 7, the analysis screen as a screen configuration displayed on the display 2 includes a PT1 displaying a voice waveform and an analysis mode, a PT2 displaying a pattern analysis, a PT3 displaying an intonation analysis, and the like. There are three types.
FIG. 8 shows a voice power analysis (hereinafter referred to as accent analysis) additionally displayed on the PT2.

In the upper half of PT1, / a /, / b /, / c /
Of the voice waveform, and the analysis mode “LOW” is illustrated in the lower half. There are "LOW" and "HIGH" in the analysis modes, which define the frequency range for searching the window size and pitch in the high-speed Fourier transform according to the height of the learner's voice. PT2 is a pattern analysis of the area enclosed by the cursors 80 and 81 of PT1, and the figure 82 is a cross-sectional view of the spectrum on the line of the cursor 83 on the left side of the cursors 83 and 84 (the size of the spectrum indicates the length of the line). It is a figure shown in FIG. The triangle mark 87 is a marker when the learner searches for the first formant and the triangle mark 88 is for searching the second formant. PT
3 shows PT1 whole intonation analysis.

Any one of the above PT1 ~ PT3 is SCR1 key 11 and SCR2
Display 2 calculated by operating key 12
It is displayed in SCREEN1 in the upper half of the upper part and SCREEN2 in the lower half.

The waveform of PT1 and the analysis mode "LOW" are set in SCREEN1 of the display 2 shown in FIG. 6, and the waveform of PT1 and the analysis mode "HIGH" are set in SCREEN2.

Next, the processing relationship in which the analysis screens shown in FIGS. 6 to 8 are displayed by the SCR1 key 11 and the SCR2 key 12 will be described with reference to the outline of the basic flowchart of FIG.

When the power switch 76 is turned on, the title image information is stored in the image memory of the SCR1 key 11 and the SCR2 key 12 in flow 100, and the title is displayed on the display 2. Flow 10
At 1, the pressing of the BEGIN key 11-1 on the SCR1 key 11 is detected to start the training. Pressing SCR1 key 11 or SCR2 key 12 is detected in flows 102 and 103. These flows are waiting for key input, and even when the process is completed and the process returns via L100, the process also comes to this flow and waits for the next key input.
If SCR1 key 11 is pressed, END key 11-2 in flow 104
Detects whether or not was pressed. If the END key 11-2 is pressed, the flow returns to the title image of flow 100, otherwise, in flow 105, the control parameter for the SCR1 key
Set the flag. Similarly, when the pressing of the SCR2 key 12 is detected in the flow 103, the control parameter flag for the SCR2 key is set in the flow 106. By setting this flag, the subsequent processing is performed by the memory 14 for each SCR1 key 11 and SCR2 key 12.
Control parameters and flags are read and written from the information set on the memory.

In the control parameter / flag memory, there are address pointers, "LOW" and "HIGH" analysis modes, and the associated window length of high-speed Fourier transform and frequency range information for pitch extraction, cursor position information, analysis screen (Fig. 7). PT1 to PT3) and associated flags (FSCR, FCL1 to 3) and other information necessary for processing the SCR1 key 11 and SCR2 key 12 are stored. If it is detected in step 107 that the WAVE MODE key 11-7 or 12-7 has been pressed, in step 108 the flag FCL1
Whether is "1" or "0". The flag FCL1 is a flag indicating whether or not the analysis screen PT1 up to the previous time is stored in the disk 15, and if it is “1”, it is not stored.
In 109, newly execute processing, display analysis screen PT1, and
Set FCL1 to "0" at 110 and record the processing result at flow 111.
Store in 15. If it is determined to be "0" in the flow 108, the processing result up to the previous time is stored in the disk 15, so this is read and displayed in the flow 112. Flow 109,112 (PL
AY) indicates that when the PLAY key 11-12 or 12-12 is pressed during these processes, the process is temporarily stopped and the contents of the audio memory are reproduced. The meaning of (PLAY) has the same meaning thereafter. Flag FSCR is set to "1" in flow 113
To The flag FSCR indicates the analysis screen being displayed, and indicates "1" for PT1, "2" for PT2, and "3" for PT3. With the above, the processing is completed, and L100
After that, the process waits for the next key input in flows 102 and 103.

Next, in PATTERN key 11-
When the pressure of 8 or 12-8 is detected, the flow 115 determines whether the flag FCL2 is "1" or "0". The flag FCL2 is a flag indicating whether or not the analysis screen PT2 up to the previous time is stored in the disk 15. After that, the process up to L100 is
Since the flow is the same as that for the WAVE MODE key 11-3 or 12-3, its explanation is omitted. Next, in flow 116, INTONATION key 11-10 for training intonation
Alternatively, when the press of the 12-10 key is detected, it is determined in the flow 117 whether the flag FCL3 is "1" or "0". The flag FCL3 is a flag indicating whether or not the analysis screen PT3 up to the previous time is stored in the disk 15. After that, the process up to L100 is similar to the above.

When the pressing of the RECORD key 11-11 or 12-11 is detected in the flow 118, the previous analysis screen PT1 is read from the disk 15 and displayed on the display 2 in the flow 119. This is a caution as to whether or not the previously recorded one can be erased. When you press the STOP key 11-13 or 12-13, you can exit to L100. Recording is executed in the flow 120, but first, recording in the voice rising memory is started. If the STOP key 11-13 or 12-13 is pressed before voice input rises,
Since there is no input, it is considered that the recording has been canceled, and the flow reaches the L100 through the flow 121 without erasing the contents of the voice memory up to the previous time. When recording is completed or after the voice input rises, press the STOP key 11-
When 13 or 12-13 is pressed, new voice data is written in the voice memory. After that, the content of the voice rising memory in which the rising portion of the voice input is recorded is transferred to the voice memory, and the flow proceeds to the flow 122. Details of the enclosed portion I will be described with reference to FIG. Since the voice memory has been refreshed, in order to invalidate all the analysis screens up to the previous time stored in the disk 15, in the flow 122, FCL1 to 3 are all set to "1" and a new execution process is performed, and then the flow 10
Go to 8. Then, the analysis screen PT1 of the newly input voice data is displayed through the flows 108, 109, 110, 111, 113, and the process reaches L100. Therefore, the teacher or the learner sees the voice waveform input by himself / herself, and as shown in FIG.
When is entered between the dotted lines LV and LV, the RECORD key 11-11 or 12-11 is pressed again to perform the processing after the flow 119.

When the press of the ACCENT key 11-9 or 12-9 is detected in the flow 123, the analysis screen displayed in the flows 124 and 125 is PT1.
It is determined whether (FSCR = 1) or PT2 (FSCR = 2). In the case of PT2, accent analysis is added by flow 126 as shown in FIG. 8, and in the case of PT1, flow 127.
To add accent analysis (power) or to display accent analysis instead of analysis mode, leading to L100.

To save the contents of the audio memory as a file, press the SAVE key 11-6 or 12-6. At this time, the file name is input by the alphanumeric key 10 of the dedicated keyboard 3 in flow 130 through flow 128 and 129, and in flow 131 S
AVE is executed, the contents of the audio memory are stored in the disk 15, and the flow goes to 132. If flow 130, STOP key 11-
When 13 or 12-13 is pressed, it is considered that filing is canceled, and the flow directly goes to the flow 132. In the flow 132, the analysis screen before displaying the file index is returned.

On the contrary, to read the contents of the voice memory from the saved file, press the LOAD key 11-4 or 12-4.

The file name is input by the alphanumeric key 10 in flow 135 through flows 133 and 134, LOAD is executed in flow 136, and the contents of the audio memory are read from the disk 15. In this case as well, since the voice memory is refreshed, in order to analyze the voice again and store it in the disk 15, all the FCL1 to FCL1 to 3 are set to "1" in the flow 137, and the flow 108 is displayed, and a new voice data analysis screen PT1 is displayed. Then it reaches L100.

If the STOP key 11-13 or 12-13 is pressed in the flow 135, it is considered that the reading of the file is canceled, and in the flow 138, the analysis screen before displaying the file index is returned to L100.

When the press of the MODE key 11-3 or 12-3 is detected in flow 139, the analysis screen displayed in flow 140 is displayed as PT1 (FSCR =
1) Determine if it is. When the analysis screen PT1 is displayed, the analysis mode is corrected in the flow 141.
That is, “LOW” → “HIGH” or “HIGH” → “LO
Change the display to "W". Along with this, the window length for high-speed Fourier transform and the frequency range information for pitch extraction in the control parameter / flag memory of the memory 14 are changed.

Therefore, it is necessary to recalculate the analysis screens PT2 and PT3, and in the flow 142, the flags FCL2 and FCL3 are set to "1", and the analysis processing is newly performed toward the L100.

Move horizontally in flow 143 Cursor keys 11-15 ₁ or 12
When the pressure of -15 ₁ is detected, the left cursor 80 (FIG. 7) is moved in the flow 144, for example. The position of the moved cursor 80 rewrites the cursor position information in the control parameter / flag memory. In the analysis screen PT1, since the pattern sandwiched between the left and right cursors 80 and 81 is subjected to pattern analysis, it is necessary to consider performing a new analysis process by moving the cursor 80. In the analysis screen PT3, since only the portion of the sound sandwiched by the left and right cursors 85 and 86 is reproduced, even if the cursors 85 and 86 are moved, new analysis processing is not required. Like the analysis screen PT2, the analysis screen PT2 only reproduces the sound between the cursors 83 and 84, so that the analysis process need not be newly performed. Therefore, the flow 145 determines whether the analysis screen is PT1 (FSCR = 1) and the flow 146
Set the flag FCL2 to "1" and proceed to the L100, taking into consideration a new analysis process.

Move horizontally in flow 147 Cursor keys 11-15 ₂ or 12
Upon detecting the pressing of -15 ₂ executes the movement of the flow 148, for example, the right of the cursor 81 (Figure 7). Left cursor 80
Since the cursor position information is rewritten like
Go to 145. The stars of flow 144 and 148 are ☆ keys 11-17, 12-
17 is moved horizontally Cursor keys 11-15 ₁ or 11-15 ₂ , 12-1
5 ₁ or 12-15 _{2 When} pressed simultaneously, cursors 80, 81
Indicates that it moves at high speed. The asterisk has the same meaning thereafter.

Move vertically in flow 149 Cursor keys 11-16 ₁ , 12-16 ₁
When the press of is detected, it is determined whether the analysis screen being displayed in flow 150 is PT2 (FSCR = 2). This key is shown in Figure 7 as a triangle 87 pointing to the first formant.
Moves. Therefore, in the flow 151, the first formant designating triangular mark 87 is moved only when the analysis screen PT2 is displayed.

Move vertically in flow 152 Cursor keys 11-16 ₂ , 12-16 ₂
Even when the press of is detected, the cursor keys 11-16 ₁ , 12-16 ₁
Similarly to the above, the case of the analysis screen PT2 is selected in the flow 153, and the triangle mark 88 indicating the second formant in FIG. 7 is moved in the flow 154.

When the press of the PRINT keys 11-5 and 12-5 is detected in flow 155, a hard copy of the display screen is printed in flow 156. If the STOP key 11-13, 12-13 is pressed midway, printing is stopped.

When the press of the PLAY keys 11-12 and 12-12 is detected in flow 157, the voice memory is reproduced in flow 158.

Although not shown in FIG. 9, HOME key 11-14, 12-1
When 4 is pressed, the cursors 80 and 81, 83 and 84, 85, 86 move to the left and right ends, respectively, and the triangular marks 87 and 88 move to the lower end.

Next, the recording operation (box I), intonation analysis operation (box II), and pattern analysis operation (box IV) shown in the basic flowchart of FIG. 9 will be described with reference to a detailed flowchart.

Recording operation The voice memory is, for example, voice as shown in Fig. 10.
59532 pieces of voice data can be stored in (0) to voice (59531). Analog voice data is a 12KHz main oscillator 4
In accordance with 2, digital data D _{2 to} D are converted by the AD converter 37.
_Since it is converted into ₁₃ (12 bits) and stored in the voice memory, the time length per voice data is 0.083 milliseconds. Therefore, in the voice memory, 59,532 units are about 4.96.
It can store voice data of a length of seconds.

The number of 59532 is related to the screen display. On the display 2 shown in FIG. 6, an X that can display a waveform is displayed.
The axial range is 1023−120−1 = 902 dots, so when displaying the contents of the voice memory in waveform, 59532/902 = 66 to 6
It is sufficient to display 6 voice data as 1 dot. There is a method to connect this by searching for the maximum value and the minimum value out of 66 (see Japanese Patent Application No. 61-3037 for details).
See the specification of No. 72).

The voice rising memory has a role as a buffer for determining whether or not the voice input has risen sufficiently, and as shown in FIG. 10, for example, 40 of vrise (0) to vrise (4095).
It has a memory capacity of 96 and can store audio data with a length of about 341 milliseconds.

The relationship between the voice rising memory and the voice memory will be described with reference to FIG.

First, in the flow 200, the output port 36 (OUT
When the output terminal REC of 3) is set to “H”, the electronic switches 40,71
Turns on. Input terminal MI selected by input selector 16
Any one input signal (audio signal) of C, AUX1 and AUX2
Input to the AD converter 37 via the electronic switch 40, the anti-aliasing filter 39 and the summing amplifier 38. The input level drives the level indicator 72 via the electronic switch 71 to turn on the LED 74 of the level meter 73 according to the input level. Check the audio level from now on.

In the flow 201, the voice rising memory pointer r = 0. When the output terminal Q ₁ of the output port 34 (OUT1) is set to “H” in the flow 202, the 3 Hz output of the frequency divider 64 passes through the AND circuit 62, passes through the OR circuit 63, and the LED 61 _{1 of the} recording time indicator 60.
Flashes. Thereby, the user confirms that the recording operation is started. STOP key 11-13 or 12-13 in flow 203
Detects whether or not was pressed. ST before the voice rises
When the OP key 11-13 or 12-13 is pressed, since the recording is canceled, the recording operation is ended. The flows 204 and 205 detect the 12 KHz signal of the main oscillator 42 input to the terminal D ₁ of the input port 33 (IN1). A- of A-D converter 37
Since the timing of D conversion is executed by the output of the main oscillator 42 as shown in FIG. 12, after detecting “L” of the output of the main oscillator 42 in Flow 204, “H” is detected in Flow 205. Immediately after that, the converted data may be taken into the input port 33. The conversion data D _{2 to} D ₁₃ input to the input port 33 in the flow 206 are stored in vrise (r) of the voice rising memory. In step 207, the pointer is incremented, and in steps 208 and 209, the pointer is circulated. In flow 210, it is determined whether or not the converted data D _{2 to} D ₁₃ of the audio signal is a certain value or more, that is, whether the audio data has risen. Before rising, it goes around the L200 and goes to the flow 203, and 4096 vrise (0), vrise (1), vrise of the voice rising memory
(2)… Converted data D _{2 to} D ₁₃ to vrise (4095) are shown in Fig. 10.
It is written cyclically as shown in FLOW1. If the conversion data D _{2 to} D _{13 of the} audio signal exceeds a certain value, proceed to L201,
As shown by FLOW2 in FIG. 10, the conversion data is continuously written from the 4097th voice (4096) of the voice memory.

In flow 211, the voice memory pointer v is set to 4096, and in flow 212 it is detected whether the STOP key 11-13 or 12-13 is pressed. When the STOP key 11-13 or 12-13 is pressed during recording, the writing to the voice memory is stopped and the process goes to L202.
As in the previous case, the flows 213 and 214 detect "L" and "H" of the terminal D ₁ of the input port 33, and the flow 215 stores the conversion data D _{2 to} D ₁₃ in the voice memory voice (v). In the flow 216, the pointer (v) is incremented, and in the flow 217, the LEDs 61 ₂ to 61 _P of the recording time indicator 60 are turned on according to the count-up of the pointer (v), and the progress of the recording time to the voice memory is progressed. To inform the user. That is, if the lighting is advanced every 500 milliseconds, 500 / 0.0
Every time v counts up 6000 from 83 = 6000, output port 34 (OUT1) output terminals Q ₂ → Q ₃ → Q ₄ may be set to “H” in that order. For example, if v is "H" from Q ₁ to Q ₂ beyond 6000 moves, the AND circuit 62 is the output of the OR circuit 63 turns off Q ₂
Causes the LED 61 ₁ that has changed to “H” and blinking to turn on. When v goes over 12,000 and "H" moves from Q ₂ to Q ₃ ,
The output of the OR circuit 63 becomes “L”, the LED 61 ₁ goes off, and the second LED 61 ₂ goes on. v is if the movement is "H" to Q ₄ 2 th LED 61 ₂ is turned off from Q ₃ exceeds 18000, the third LE
D61 ₃ Lights up. The same applies hereinafter. With flow 218 v> 59531
Then, the end of recording is detected in flow 218. If it is not past the end of the voice memory, go back around L203 to return to voice (4096), voice (4097), voice (409
8), ... Write voice data to voice (59531) one after another. When the recording is complete, the contents of the voice rise memory are transferred to the voice memory to complete the storage in the voice memory. That is, when the flow 218 ends, the voice memory pointer v = 0 is set in a flow 219 via L202. Flow 220
Then, only one content of the voice rising memory vrice (r) is transferred to the memory. In the flow 221, the two pointers v and r are incremented, and in the flows 222 and 223, the voice rising pointer r is circulated. The flow 224 detects the end of transfer. L204 unless all the contents of the voice start-up memory are transferred
It goes around and returns, and transfers are made one after another. When the transfer is completed, the voice memory is complete, so the flow advances to L205 to proceed to the waveform display of the analysis screen PT1 via the flow 122 of FIG. The transfer of the voice rising memory to the voice memory in the flow 220 is sequentially transferred to the voice memory from the content which is advanced by one from that when the converted data D _{2 to} D ₁₃ is determined to be a certain value or more in the flow 210. This is because the voice rising memory pointer r is incremented in the flow 207. For example, when the conversion data D _{2 to} D ₁₃ of the vrise (1525) of the voice rising memory is detected to be a certain value or more in the flow 210, the elapsed time of the contents of the voice rising memory is vrise (1525).
The next vrise (1526) is the oldest, and the vrise (152
7), vrise (1528), vrise (1529), and so on through vrise (15
25) is the newest. Therefore, voice vrise (1526)
(0), vrise (1527) to voice (1), vrise (1528) to v
By transferring vrise (1525) to voice (4095) via oice (2), ..., the recording time of the voice memory becomes complete. Applying to Fig. 11 and considering it, the flow 20
In step 6, the converted data D _{2 to} D ₁₃ are stored in the rise (1525), and in the flow 207, r = 1526 and the flow 210 determines that the value is equal to or more than a certain value, and the process exits to L201. Therefore, in flow 220, first vrise
(1526) is transferred to voice (0), and the two pointers are incremented in flow 221.
Transfer 7) to voice (1). The same applies hereinafter.

As described above, by providing the voice start-up memory, recording to the voice memory automatically starts when voice input rises, and the stop key 11-13, 12
When -13 is pressed, the recording can be canceled and the previous contents of the voice memory will not be erased. Also, since the voice data before rising is stored in the voice rising memory for about 340 milliseconds, there is no concern that recording will start halfway.

Intonation analysis operation Intonation analysis operation is performed in the next box portion II of the flow 117 of the basic flowchart shown in FIG.
The details are shown in the flowchart of FIG.

First, in flow 300, output port P of output port 36 (OUT3)
When I is set to "H", the electronic switch 70 is turned on. D-
The output of the A converter 66 is a low-pass filter 69, an electronic switch
It is input to the A / D converter 37 via 70 and the addition amplifier 38.

The flow memory 301 sets the voice memory pointer v = 0, the working memory pointer w = 0, and the counter c = 3, and then the flows 302 and 303 detect the 12 KHz output of the main oscillator 42 input to the terminal D ₁ of the input port 33 (IN1). To do. When the terminal D changes from "L" to "H", the flow of voice memory voice
The audio data is output from (r) to the output port 35 (OUT2). Since the timing of this output depends on the judgment timing of the PC 1, jitter is not included in the fixed time, so the latch 65 synchronizes with the rising edge of the output of 12 KHz, and then the DA converter. At 66, it is converted to an analog audio signal (see Fig. 12). The analog voice signal is reduced by the low-pass filter 69 to eliminate high frequencies that are unnecessary for pitch extraction, and then converted into a digital voice signal again by the AD converter 37. In flow 305, this converted data D ₂ ~ D
_The voice memory pointer is incremented in a flow 306 in which ₁₃ is fetched and stored in w mem (W) of the working memory, and the counter c is decremented in a flow 307. In the flows 308 to 310, the working memory pointer w is not incremented until c = 0, as can be seen from the flow. In other words, every time the flow 305 is passed, the conversion data D _{2 to} D ₁₃ are transferred to the working memory w mem.
It is incremented after overwriting (W) three times.
Therefore, the remaining conversion data is written every third time. This means that the sampling frequency of the working memory w mem (W) has dropped to 12 KHz / 3 = 4 KHz. The low-pass filter 69 reduces the bandwidth of the analog audio signal.
Since it was narrowed down to about 900Hz, this is sufficient. Therefore, the working memory capacity in this case is 1/3 of the voice memory.
(From 59532/3 = 19844 w mem (0) to w mem (19843))
The above is sufficient. In flow 311, the end of reproduction of the audio memory is detected. As long as the last part of the voice memory is not exceeded,
Turn around L300 to return, and convert data to working memory D ₂ ~
Write D ₁₃ one after another.

Based on the contents of the working memory storing the conversion data D _{2 to} D ₁₃ as described above, the pitch extraction analysis is performed in the flow 312, and the result is the pitch value data pitch.
It is stored in the buffer memory as (i). Now, it is assumed that one pitch value data is obtained for every 88 converted data in the working memory. That is, w mem (0) to w mem (87)
At pitch (0), w mem (88) ~ w mem (175) at pitch
(1), and so on. Since the number of working memories is 19844, 1984/88 = 225.5, so that pitch value data pitch (0) to pitch (224) are calculated.

The method of calculating the pitch value data is not the main purpose of the present invention, and therefore will not be described in detail (see the specification of Japanese Patent Application No. 61-303772).

After the flow 313, the intermittent reproduction operation is executed to intermittently reproduce the display of the intonation and the sounded sound to clarify the correspondence between the display of the intonation and the sounded sound.

An example of the intermittent playback operation will be described with reference to FIG. When the intermittent reproduction operation is performed for the intonation display shown in the sections k _{1 to} k ₇ in FIG. 14 (A), the pitch value data
k ₂ , k ₄ , k ₆ where pitch (i) can be extracted and graph can be displayed
Is a section of a voiced sound close to a line spectrum structure in which a relatively stable waveform such as a vowel repeats. For this, pitch (i) = 0 cannot be extracted and graph display is not possible k ₁ , k ₃ , k ₅ and k ₇ are unvoiced intervals having a continuous spectral structure with silent intervals or random waveforms.

The learner must look at FIG. 14 (A) to grasp the defects of his / her intonation, but the learner may be confused by not knowing the correspondence between the pronunciation content and the intonation display of FIG. 4 (A). Intonation in particular deals with long sentences pronounced fluently, so considerable training is required to grasp the correspondence even if k _{1 to} k ₇ are played continuously. The intermittent playback operation solves such inconveniences as
The intonation display and the playback are intermittently advanced so that the correspondence between the pronunciation content and the intonation display can be easily grasped. That is, as shown in FIG. 14 (B), after the intonation display up to k ₁ and k ₂ is displayed, the reproduction is performed and after waiting for 500 milliseconds, as shown in FIG. 4 (C), a section of k ₃ k ₄ is displayed. The intonation of is displayed and then played. And so on 500
The time shifts from FIG. 4 (D) to FIG. 4 (E) intermittently with a waiting time of about milliseconds. The 500 ms waiting time gives an impression that the correspondence is easy to grasp and the whole flow of pronunciation is easily grasped. As shown in FIG. 14 (A), see there is no problem when the boundary p _1, p _2, p ₃ intermittent point the rear of the voiced and bring the front, often rising portion of the speech It is not preferable because the unvoiced sound generated is often inaudible at the start of intermittent reproduction, and the information on the rising portion of the voice, which is important for understanding words, is often omitted.

In flow 313, pitch data pointer i = 0, voice memory pointer v = 0, and voice memory second pointer dv = 0. In flow 314, it is determined whether the pitch value data pitch (i) is voiced sound, unvoiced sound, or unvoiced sound. If pitch (i)> 0, it is a voiced sound, and therefore flow 315 displays pitch (i) on the screen and goes to flow 316. When pitch (i) = 0, there is no sound or unvoiced sound, so the flow goes directly to the flow 316. In the flow 316, the voice memory pointer v used at the time of re-correction is corrected. The pitch data pointer i is incremented for each 88 working memories dropped to a sampling frequency of 1/3 of the voice memory to obtain pitch value data. Therefore, it is necessary to advance the voice memory pointer v by 88 × 3. At flow 317, the intermittent points (p ₁ , p ₂ , p _{3 in} FIG. 14) are detected. If the pitch value data pitch (i) currently being processed is a voiced sound and the pitch (i + 1) advanced by one is silent or unvoiced, it is judged to be an interruption point and the flow proceeds to the flow 318. Since it is necessary to display the pitch value data in a graph continuously, the flow goes to the flow 321. In flow 321, the pitch data pointer i is incremented, and in flow 322, until the last part i = 224 of the pitch value data is passed, the process goes around L301 and returns to flow 314. When it is determined in the flow 317 that it is the intermittent point, the display of the pitch value data up to the intermittent point has already been completed, so the voice up to the intermittent point is reproduced in the flow 318. That is, the voice data from voice (dv) to voice (V) in the voice memory is output to the output port 35 (OUT2), and is converted into an analog voice signal by the DA converter 66. At this time, the output terminal PL of the output port 36 (OUT3) becomes "H" and the electronic switch 67 is turned on. Therefore, the analog audio signal is transferred to the electronic switch 67, the addition amplifier 26, the low-pass filter 27 and the power amplifier 29. Then drive speakers and headphones from the output terminals SP.OUT, PHONE1 and PHONE2. Referring to FIG. 14, when the flow 318 first comes, as shown in FIG. 14 (B), the intonation of k ₁ k ₂ is displayed, and the position of v = p ₁ is designated with dv = 0. Therefore, the sound corresponding to the intonation of the section k ₁ k ₂ is reproduced. Flow 319 to 500
After waiting for milliseconds, flow 320 sets dv = p ₁ . When the player turns around L301 and then comes to flow 318, as shown in FIG. 14 (C), the intonation at k ₃ and k ₄ is displayed, and the position of v = p ₂ is indicated by dv = p _1. Section k ₃ k ₄
The sound corresponding to the display of the intonation of is played. Similarly, the next intonation of k ₅ and k ₆ is displayed as shown in FIG. 4 (D), and the position of v = p ₃ is indicated by dv = p ₂ , so the intonation display of the section k ₅ k ₆ is displayed. The sound corresponding to is played. The flow 323 is for reproducing the section k _{7 in} FIG. 4 (E), and since dv = p ₃ ,
The voice data voice (p ₃ ) to the last voice (59531) are played.

In the boxed portion III in FIG. 9, the contents of the pitch extraction of the buffer memory up to the previous time may be read again from the disk and the intermittent reproduction operation after the direct flow 313 may be performed.

Pattern Analysis Operation The pattern analysis operation is performed in the next boxed portion IV of the flow 115 of the basic flowchart shown in FIG. 9, the details of which are shown in FIG.

First, in Flow 400, output terminal S of output port 36 (OUT3)
When G is set to "H", the electronic switch 41 is turned on and D-
The output of the A converter 66 is a high-pass filter 68, an electronic switch
A circuit for inputting to the AD converter 37 via 41, the anti-aliasing filter 39, and the adding amplifier 38 is formed. flow
In 401, the voice memory pointer v = v ₁ and the working memory pointer w = 0 are set. As described in FIG. 7, the analysis screen PT2 which is the pattern analysis is the cursor 80-81 of the analysis screen PT1.
This is a pattern analysis of the area surrounded by, and the positions of the cursors 80 and 81 in the voice memory are voice (v ₁ ) and
voice (v ₂ ).

The flows 402 and 403 detect the 12 KHz output of the main oscillator 42 input to the terminal D ₁ of the input port 33 (IN1), and the flow 404
Outputs voice data of voice (V) in voice memory at port 3
The analog audio signal is output from the DA converter 66 by outputting to 5 (OUT2). The detailed operation of the flows 402 to 404 is the same as that described in the flows 302 to 304. The high-pass filter 68 emphasizes the high-frequency components necessary for pattern analysis until the analog voice signal is input to the A / D converter 37, and the anti-aliasing filter 39 further components other than the voice band frequency. Is removed, and the AD converter 37
Is converted into a digital voice signal again. In the flow 405, the converted data D _{2 to} D ₁₃ are fetched and stored in the working memory w mem (W). The pointers are incremented in the flows 406 and 407, and in the flow 408, the converted data D _{2 to} D ₁₃ are written in the working memory one after another while returning by going back to the L400 before returning from the cursor range on the voice memory.

Based on the contents of the working memory stored as described above, a high speed Fourier transform is performed in flow 409 to calculate a power spectrum value, which is converted into a grayscale number of grayscale and stored in the buffer memory. About 7 gradations are sufficient, so the numerical values are 1 to 7. This gradation number data is stored when the processing result is stored in the disc.

Further, the grayscale data is selected based on the gradation number data of 1 to 7 and stored in the image memory. As the grayscale data,
There are bit pattern segments as shown in Fig. 16 ("H" for dark areas, "L" for thin areas). Eventually, the bit pattern segments shown in FIG. 16 are stored in the image memory so as to lay tiles to form a grayscale image such as the analysis screen PT2.

Reproduction Operation The reproduction operation is performed according to the flowchart shown in FIG.

First, in flow 500, output port P of output port 36 (OUT3)
When L is set to "H", electronic switch 67 is turned on and D-
The output of the A converter 66 is a summing amplifier 26 and a low pass filter 27.
And input to the power amplifier 29. Similar to the pattern analysis, the voice memory pointer v = v ₁ is set in the flow 501, and the flow 5
02 and 503 detect the 12 KHz output of the main oscillator 42 at the terminal D ₁ of the input port 38 (IN1), and the flow 504 outputs the contents of voice (V) of the voice memory to the output port 35 (OUT2) and DA The converter 66 outputs an analog audio signal.
The pointer v is incremented in flow 505, and flow 506
In one not out of the range of the cursor on the voice memory (v ₁ ~v ₂₎ returns around the L500 to repeat the reproduction operation. As described above, since the reproduction operation is limited to the cursor, it is possible to reproduce an arbitrary section on the recorded waveform and grasp the correspondence between the waveform and the pronunciation. If the reproducing operation is performed on PT2, which is the pattern analysis as shown in FIG. 7, the distance v ₀ between cursors v ₀ = v ₂ −v ₁ is the voice memory of cursors 83 and 84 on PT2. Position voice (v ₁ ), voice (v ₂ )
Will be required from. Therefore, it is possible to reproduce the arbitrary section on the pattern and grasp the correspondence between the pattern and the pronunciation.

(Effects of the Invention) Since the present invention is configured as described above, it has the effects described below. According to the pronunciation training device of claim 1,
Since the voice analysis results of the teacher and the learner can be respectively displayed on the display as a pair of screens, the teacher can immediately point out the learner's pronunciation error and the learner can easily understand. In addition, the teacher and the learner can efficiently select the desired voice analysis result by each key, which is efficient. Then, when training intonation, it is possible to clearly grasp the correspondence between the display of intonation and the pronunciation sound even in a long sentence that is pronounced fluently. According to the pronunciation training apparatus of the second aspect, when the voice rises, the recording in the voice memory is automatically started, and the recording does not start halfway. According to the invention described in claim 3, it is possible to determine the loudness of the pronunciation of the learner and adjust the recording level of the audio signal of other audio signal generating means such as a tape recorder to a desired value.

[Brief description of drawings]

FIG. 1 is a perspective view showing the external appearance of an embodiment of the present invention, FIG. 2 is a plan view of the dedicated keyboard, FIG. 3 is a diagram of the personal computer, and FIGS. Front and rear views of the unit, FIG. 5 is a circuit diagram of its external circuit, FIG. 6 is a front view showing an example of its display, and FIG. 7 is three examples of analysis screens displayed on the display. FIG. 8, FIG. 8 is a diagram showing a modified example thereof, FIG. 9 is a basic flowchart thereof, FIG. 10 is an explanatory diagram of the memory thereof, FIG. 11 is a flowchart of its main part, and FIG. 12 is its main part. Operation explanatory diagram, FIG. 13 is a flowchart of other main parts, FIGS. 14 (A) to (E) are operation explanatory diagrams thereof, and FIGS. 15 and 17 are flowcharts of other main parts, FIG. 16 respectively. FIG. 6 is an explanatory diagram of correspondence between gradation numbers of gradations and gradation data. 1 ... Personal computer 2 ... Display, 3 ... dedicated keyboard 4 ... dedicated unit, 11 ... teacher key 12 ... learner key

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Katsuhiro Hashimoto 200 Terashimacho Hamamatsu City, Shizuoka Prefecture Kawai Musical Instrument Co., Ltd. (72) Inventor Michiko Nihei 200 Terashimacho Hamamatsu City Shizuoka Prefecture Kawai Musical Instrument Co., Ltd. (56) References JP 61-121077 (JP, A) JP 61-6730 (JP, A) JP 61-6732 (JP, A) JP 59-214880 (JP, A) JP-A-59-12479 (JP, A) JP-A-57-96377 (JP, A)

Claims (3)

[Claims] 1. A voice signal generating means, a voice analyzing means for analyzing a voice associated with pronunciation from a voice signal, a memory for storing at least voice data and a voice analysis result, and a pair of an effective screen upper half and lower half. A display having a screen configuration, two sets of key groups consisting of a plurality of operation keys in which at least the same training items are arranged at the same relative positions corresponding to the respective screens, and a pair of keys on the display are operated by operating the key groups. In a pronunciation training apparatus provided with means for controlling and independently displaying the screen configuration and comparing and displaying the voice analysis results, the pitch value data written in the memory is sequentially read out, and when the pitch value data is voiced, the pitch value data is read. The value data is displayed on the display, the read pitch value data is voiced, and the next read pitch value data is silent or When it becomes a voice, an intonation analysis is performed in which the voice corresponding to the pitch value data of the voiced sound read continuously until then is reproduced, and after a lapse of a predetermined time, the series of operations is performed again for the remaining pitch value data. A pronunciation training device comprising an operating means. 2. The memory includes a voice memory for storing voice data and a voice rising memory, wherein a write operation is circulated to the voice rising memory to write a rising portion of the voice data to the voice rising memory, When the voice data exceeds a certain value, the writing operation is switched to the voice memory, the voice data is written following the unstored area corresponding to the voice rise memory of the voice memory, and after the writing is finished, the voice rise memory is written. 2. The pronunciation training apparatus according to claim 1, further comprising recording operation means for transferring the written voice data to an unstored area of the voice memory. 3. The pronunciation training apparatus according to claim 1, wherein a recording level adjusting volume is provided in the signal transmission circuit of the audio signal generating means other than the microphone.