JPH10207349A - Card type language learning machine and voice comparing system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To objectively and exactly judge the uttering of learning person. SOLUTION: When a magnetic data reading mechanism part 2 reads the model uttering of magnetic tape 1a, a model uttering feature extracting means 3 extracts the respective features of accent and intonation from this read model uttering. On the other hand, the uttering of learning person is collected from a microphone 4. A learning person uttering feature extracting means 5 extracts the respective features of accent and intonation in the uttering of learning person. A pattern matching means 6 performs pattern matching between the respective extracted features in model uttering and learning person uttering. While using three kinds of LED 8a, 8b and 8c of display panel, for example, a matching display means 7 displays the result of pattern matching.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a card type language learning machine for performing language learning and a voice comparison system.
In particular, the present invention relates to a card-type language learning machine and a voice comparison system having a function of comparing a student's utterance with a model utterance.

[0002]

2. Description of the Related Art Conventionally, as a card type language learning machine, a magnetic card called a program card is conveyed along a rail as shown in Utility Model Registration No. 3023873, and a magnetic tape attached to the magnetic card is used. There is a device that reads a model utterance recorded in a speaker and outputs it from a speaker.

[0003] In the learning procedure by such a card-type language learning machine, a model utterance recorded on a magnetic card is listened to by a speaker. The learner himself speaks into the microphone and records it on a magnetic card.

[0004] By reproducing the magnetic card, the model utterance is compared with the utterance of the learner himself. It is common. Then, the learner practices by repeating ~ to make his or her utterance closer to the model utterance.

[0005]

However, in the conventional card type language learning machine, since the learner only compares his / her own utterance with the model utterance, it is not possible to objectively and accurately judge the difference between the two. could not.

[0006] On the other hand, some language teaching software for general personal computers displays a pronunciation waveform on a CRT or the like so that a model utterance can be compared with a student's utterance.
However, utterance elements include accent and intonation as well as pronunciation. That is, when these elements are harmonized, a correct utterance is obtained. Therefore, it is not enough to compare the waveforms of the sounds as in the related art.

The present invention has been made in view of the above points, and has as its object to provide a card-type language learning machine and a voice comparison system capable of objectively and accurately determining the utterance of a learner. And

[0008]

According to the present invention, in order to solve the above-mentioned problems, in a card-type language learning machine for performing language learning using a magnetic card, an exemplary utterance data area in which an exemplary utterance is recorded is stored. A magnetic data reading mechanism for reading the magnetic data of the magnetic card, a microphone for collecting the learner's utterance, an exemplary utterance feature extracting means for extracting each characteristic of accent and intonation of the exemplary utterance of the magnetic card, Learner utterance feature extraction means for extracting each feature of the accent and intonation of the learner utterance; pattern matching means for pattern matching each of the extracted model utterance features and each of the learner utterance features; And a matting display means for displaying a result of the matching.習機 is provided.

In such a card-type language learning machine, the magnetic data reading mechanism reads magnetic data of a magnetic card and reads model speech. From the read model utterance, the model utterance feature extracting unit extracts each feature of the accent and intonation. On the other hand, learners' utterances are collected from the microphone. From the collected learner utterances, the learner utterance feature extraction means extracts each feature of the accent and intonation.

The pattern matching means performs pattern matching between each feature of the extracted model utterance and each feature of the learner utterance. Then, the matching display means displays the result of the pattern matching. Thereby, the difference between the learner utterance and the model utterance is objectively and accurately evaluated, and it can be confirmed at a glance.

[0011]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing the concept of the function of the card type language learning machine of the present embodiment. On the magnetic tape 1a of the magnetic card 1, at least magnetic data of the model utterance is recorded. The magnetic data reading mechanism 2 reads the model utterance of the magnetic tape 1a. Model utterance feature extraction means 3
Extracts the features of accent and intonation from the read model utterance. On the other hand, learners' utterances are collected from the microphone 4. The learner utterance feature extraction means 5 extracts each feature of the accent and intonation of the learner utterance.

The pattern matching means 6 performs pattern matching between each feature of the extracted model utterance and each feature of the learner utterance. The matching display means 7 uses, for example, three types of LEDs 8a, 8b, 8c of the display panel 8,
Displays the result of pattern matching. That is, if the degree of matching is bad, the LED 8a is turned on, if it is good, the LED 8b is turned on, and if it is very good, the LED 8c is turned on.
Lights up.

FIG. 2 is a perspective view showing the external configuration of the card type language learning machine of the present embodiment. The upper case 12 of the card-type language learning machine 10 includes an operation unit 13 for performing various operations described later, a sound output unit 14 for outputting a sound from a speaker 34 described later inside the case, and a transport for the magnetic card 16. A reading mechanism 15 and the like are provided.

The transport reading mechanism 15 inserts the magnetic card 16 into the rail, thereby recording and reading utterance data on the magnetic tape 16a while transporting the magnetic card 16 in the direction of the arrow in the figure.

The operation unit 13 includes a student mode button 21,
Teacher mode button 22, Record button 23, Matching switch 24, Volume 25, Microphone hole 26, Display unit 2
7 are provided. The student mode button 21 is a button that is pressed when recording or reproducing a student's utterance. When the student mode button 21 is pressed at the same time as the recording button 23, the magnetic tape 16 of the magnetic card 16 is pressed.
The utterance of the learner can be recorded in a. When only the student mode button 21 is pressed, the learner's utterance recorded on the magnetic tape 16 a of the magnetic card 16 is output from the audio output unit 14.

On the other hand, the teacher mode button 22 is a button to be pressed when listening to the model utterance recorded on the magnetic tape 16a. The matching switch 24 is a switch for performing pattern matching between the model utterance and the learner utterance. When the teacher mode button 22 is pressed with the matching switch 24 turned on and the magnetic card 16 is read, the model utterance of the magnetic tape 16a is read, and the characteristics of the model utterance are extracted by the procedure described later.
Stored in internal memory. On the other hand, when the student mode button 21 is pressed in a state where the matching switch 24 is turned on and a learner's utterance is collected from a microphone 35 to be described later through the microphone hole 26, the characteristics of the learner's utterance are described by a procedure described later. Is extracted. Then, the features of the model utterance and the learner utterance stored earlier are subjected to pattern matching by a method described later, and as a result, the similarity between the two is displayed on the display unit 27.

FIG. 3 is a diagram showing an example of the configuration of the display unit 27. The display section 27 includes, for example, display fields for three pictures, that is, a circle display field 27a, a double circle display field 27b,
A flower circle display field 27c is provided. Each display column 27a,
LEDs (not shown) are provided inside 27b and 27c, and when the LEDs are turned on, each picture is displayed. Here, as a result of the pattern matching, when the degree of matching between the model utterance and the feature of the learner utterance is low, the circle display field 27a is turned on. When the degree is medium, the double circle display field 27b is high. , The flower circle display field 27c is turned on.

FIG. 4 is a block diagram showing a hardware configuration inside the card type language learning machine 10. As shown in FIG. Control circuit 31
Is composed of, for example, a logic circuit. Control circuit 3
1 is a student mode button 21, a teacher mode button 22,
The transport reading mechanism 1 according to the state of the recording button 23, the matching switch 24, a sensor (not shown), and the like.
5 to control the operation of the entire card type language learning machine 10.

The audio processing circuit 32 is constituted by, for example, a logic circuit. The audio processing circuit 32 includes a later-described matching processing circuit, a preamplifier circuit, a memory, a frequency modulation circuit, and the like. That is, the audio processing circuit 32
The learner's utterance collected by the microphone 35 is read, and its features (accent and intonation) are extracted.
On the other hand, the audio processing circuit 32 controls the magnetic card 1 by the magnetic head 151 in response to a command from the control circuit 31.
6 to read the model utterance data and extract the features (accents and intonation) of the model utterance. And
The feature of the learner utterance and the feature of the model utterance are subjected to pattern matching, and the result is sent to the control circuit 31.

The control circuit 31 receiving the result of the pattern matching displays the similarity between the feature of the learner's utterance and the feature of the model utterance on the display unit 27. Also, the audio processing circuit 3
2 reads the model utterance data of the magnetic card 16,
It is amplified and sent to the power amplifier 33. The model utterance sent from the audio processing circuit 32 is amplified according to the degree of adjustment of the volume 25 and output to the speaker 34. Note that the learner's utterance collected by the microphone 35 is
, Or output from the speaker 34.

FIG. 5 is a block diagram showing a specific configuration of the matching processing circuit in the audio processing circuit 32. Also,
FIG. 6 is a diagram showing functions of a matching processing circuit from input of an audio signal to extraction of a feature pattern. The matching processing circuit 40 shown in FIG. 5 receives, for example, a model utterance or a learner utterance voice signal 51 having a waveform as shown in FIG. The input audio signal 51 is input to the accent detection circuit 4.
1 and sent to the intonation detection circuit 42.

The accent detection circuit 41 has a voice signal 51
Is a circuit that detects changes in accent from the rectifier,
It is composed of a smooth LPF (low-pass filter) or the like.
In the accent detection circuit 41, after rectifying the waveform of the audio signal 51, the LP having a cutoff frequency of about several tens Hz
The signal is smoothed by F to obtain a signal as shown in a graph 52 of FIG. Here, the horizontal axis of the graph 52 indicates time, and the vertical axis indicates the voltage level of the detected audio amplitude. The recognition of the accent only needs to match the strength of the voice, so it is not necessary to match the strength of the voice.
Therefore, in the extraction of the accent feature pattern, the recognition rate is improved by a method such as level compression of the voice level or modeling of the pattern.

On the other hand, the intonation detection circuit 42
This circuit detects an intonation waveform from the audio signal 51, that is, a change in the level of the audio frequency, and includes a LPF,
It is composed of a zero-crossing wave detection circuit, an F / V (frequency / voltage) conversion circuit, and the like. In the intonation detection circuit 42, a signal as shown by a graph 53 in FIG. 6 is obtained by detecting a change in the basic pitch frequency of the audio signal 51. Basic pitch frequency is 125Hz for adult men and 250H for women
about z. Here, the higher frequency, 250
Hz plus the intonation change α, (250+
α) It is detected by an LPF having a cutoff frequency of Hz.
In the graph 53, the F / V conversion is performed for the sake of processing, and the vertical axis indicates the voltage level, but the change in the voltage level corresponds to the change in the basic pitch.

In general, since the variation of intonation is small, modeling is easier than in the case of accent. The basic pitch that could not be detected when the utterance was interrupted is smoothly patterned before and after the basic pitch.

When the accent and intonation waveforms are detected in this manner, the multiplexer 43 shown in FIG.
Sends them to the A / D conversion circuit 44 in series. A / D
The conversion circuit 44 converts an analog signal into a digital signal.

The feature pattern extraction circuit 45 extracts a feature pattern from each of the detected accent and intonation waveforms. As a method for extracting the feature pattern, generally, a linear prediction coefficient, a PARCOR coefficient,
There is a method based on band filter output, zero crossing wave coefficient, energy, autocorrelation function, and the like. This embodiment can be executed by using any of these. The respective characteristic patterns of accent and intonation are shown in Tables 54 and 55 of FIG. 6, respectively.

Each characteristic pattern extracted by the characteristic pattern extraction circuit 45 shown in FIG.
Sent to That is, if the extracted feature pattern is a model utterance signal, the changeover switch 46 is switched by a command from the control circuit 31 and sent to the model utterance storage circuit 47, where it is temporarily stored. . On the other hand, if the extracted feature pattern is a signal of a learner's utterance, it is sent directly to the pattern matching circuit 48.

The pattern matching circuit 48 compares the accent and intonation of the model utterance stored in the model utterance storage circuit 47 with the accent and intonation of the learner's utterance, and calculates the degree of similarity between the two. Here, the pattern matching is D
Use P (dynamic programming) matching.

Next, a specific example of pattern matching of the present embodiment using DP matching will be described. FIG. 7 is a diagram showing a method of comparing the feature patterns of each accent of the model utterance and the learner utterance. Here, about the model utterance,
The feature pattern of Table 54 shown in FIG. 6 is used. Characteristic pattern of this exemplary utterance, as shown, are accommodated between the sampling time t ₀ ~t _12. On the other hand, learner utterance, this faster tempo than the short time interval t ₀ ~t
Fits between _nine . According to the DP matching, an error between data having values as close as possible between the model utterance and the learner utterance is calculated. However, calculation methods in which the front and rear cross and the arrows in the figure cross are prohibited.

When the error between each data is calculated by this method and the sum of their absolute values (hereinafter, referred to as “distance”) is calculated, it becomes 0 here. That is, in the example of FIG. 7, it can be seen that the accents of the model utterance and the learner utterance are very similar even if the tempo is different.

The pattern matching circuit 48 calculates the intonation between the model utterance and the learner utterance in the same manner. When the matching of each feature pattern is completed, the similarity determination result is sent to the control circuit 31.
Send to

The control circuit 31 which has received the instruction causes the display unit 27 to perform a display corresponding thereto. For example, if the average of each distance between the accent and the intonation is 9 or more, the circle display field 27a shown in FIG. 3 is turned on. If the number is 5 or more and 8 or less, the double circle display section 27b is turned on. If the number is 4 or less, the flower circle display section 27c is turned on.

When it is desired to display accents and intonations separately, display columns 27a and 27 in FIG.
A pair of LEDs having different colors may be provided for b and 27c, and each of them may be turned on according to the above-mentioned point distribution.

FIG. 8 is a flowchart showing a procedure performed by the matching processing circuit 40. [S1] It is determined whether or not the operation unit 13 of the main unit is in the teacher mode. If yes, the process proceeds to step S2, and if not, the process proceeds to step S7. [S2] The changeover switch 46 shown in FIG. 5 is switched to the model utterance storage circuit 47 side. [S3] The accent and intonation of the model utterance are detected. [S4] A / D conversion of the detected signal is performed. [S5] The accent and intonation feature patterns of the model utterance are extracted.

[S6] The extracted pattern is registered in the model utterance storage circuit 47. [S7] Set the changeover switch 46 to the pattern matching circuit 4
Switch to 8 side. [S8] The accent and intonation of the learner's utterance are detected. [S9] A / D conversion of the detected signal is performed. [S10] Each feature pattern of the accent and intonation of the learner's utterance is extracted. [S11] Pattern matching of each characteristic pattern of the model utterance and the learner utterance is performed. [S12] The similarity between the two is displayed.

As described above, in the present embodiment, the degree of similarity is displayed by comparing the accent and intonation of the model utterance and the learner's utterance, so that the difference between the learner's utterance and the model utterance is objectively determined. And can be evaluated accurately and at a glance.

In this embodiment, DP matching is used as a pattern matching method.
Even if the learner utterance and the model utterance have different tempos, accurate matching can be performed.

In this embodiment, the card type language learning machine 1
Although the example using 0 is shown, the function of the present embodiment can be executed by a normal personal computer or the like by attaching a voice input board. At this time, the pattern matching process is performed by software, and the similarity is displayed on a monitor or the like.

[0039]

As described above, according to the present invention, each feature of the accent and intonation is extracted from the model utterance, while each feature of the accent and intonation is extracted from the learner utterance collected from the microphone. Each feature of the extracted model utterance and each feature of the learner's utterance were subjected to pattern matching, and the results of the pattern matching were displayed, so that the difference between the learner's utterance and the model utterance was objectively and Accurate evaluation can be made, and it can be confirmed at a glance.

[Brief description of the drawings]

FIG. 1 is a diagram showing a concept of a function of a card-type language learning machine of the present embodiment.

FIG. 2 is a perspective view showing an external configuration of a card-type language learning machine of the present embodiment.

FIG. 3 is a diagram illustrating an example of a configuration of a display unit.

FIG. 4 is a block diagram showing a hardware configuration inside the card type language learning machine.

FIG. 5 is a block diagram showing a specific configuration of a matching processing circuit in the audio processing circuit.

FIG. 6 is a diagram illustrating functions of a matching processing circuit from input of an audio signal to extraction of a feature pattern.

FIG. 7 is a diagram showing a method of comparing feature patterns of accents of a model utterance and a learner utterance.

FIG. 8 is a flowchart illustrating a procedure performed by a matching processing circuit.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Magnetic card 2 Magnetic data reading mechanism part 3 Model utterance feature extraction means 4 Microphone 5 Learner utterance feature extraction means 6 Pattern matching means 7 Matching display means 8 Display panel 10 Card language learning machine 13 Operation part 16 Magnetic card 16a Magnetic tape 21 Student Mode Button 22 Teacher Mode Button 23 Record Button 24 Matching Switch 27 Display 31 Control Circuit 32 Audio Processing Circuit 35 Microphone 40 Matching Processing Circuit

Claims (4)

[Claims] 1. A card-type language learning machine for performing language learning using a magnetic card, comprising: a magnetic data reading mechanism for reading magnetic data of a magnetic card having a model utterance data area in which a model utterance is recorded; A microphone that collects learner utterances; a model utterance feature extraction unit that extracts each feature of accent and intonation of the model utterance of the magnetic card; a learner utterance that extracts each feature of accent and intonation of the learner utterance A card comprising: a feature extracting unit; a pattern matching unit that performs pattern matching between the extracted model utterance feature and the learner utterance feature; and a matching display unit that displays a result of the pattern matching. Expression language learning machine. 2. The card-type language learning machine according to claim 1, wherein the pattern matching is DP (Dynamic Programming) matching. 3. The card-type language learning machine according to claim 1, wherein the matching display means is configured to display a picture corresponding to the degree of similarity of the matching. 4. A voice comparison system for comparing a model utterance with a learner utterance, comprising: a model utterance data storage unit storing a model utterance; a microphone for collecting a learner utterance; and an accent of the model utterance. Model utterance feature extraction means for extracting each feature of the learner and intonation; learner utterance feature extraction means for extracting each feature of the learner's utterance accent and intonation; the extracted model utterance feature and the learner A voice comparison system comprising: a pattern matching unit that performs pattern matching with an utterance feature; and a matching display unit that displays a result of the pattern matching.