JPS60163078A - Speech training apparatus - Google Patents

Abstract

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

Description

[Detailed Description of the Invention] Industrial Application Field The present invention is a speech training method in which speech training is performed by extracting feature signals of speech accompanying speech, and comparing and displaying the feature signals and its training model on a display device. It is related to the device.

Configuration of conventional examples and their problems In recent years, due to improvements in technology for extracting features of the vocal organs for speech and advances in visual display technology, there are various speech training devices that perform speech training by extracting features and visually displaying them in real time. It has been devised.

Conventionally, in such speech training devices, a training model for all spoken words to be trained, created in advance, is loaded and displayed in a section of the screen of the display device in sequence for each spoken word. Training was performed by comparing and displaying the speaker's feature extraction data on the same time axis of the model. The utterance training model is provided for each utterance word,
This is a model that is considered to be the basic utterance of a founder, and is created by, for example, a language trainer.

In speech training, due to the sharpness of human speech, even if the model and the speaker's feature extraction data are exactly the same, if a certain degree of approximation is achieved, speech intelligibility and intelligibility will improve. .

In conventional speech training devices such as those mentioned above, one fixed data training model is provided for each spoken word, but in the initial stage of training, it is difficult to focus on which part of the complex training model to train. The problem is that it is difficult to determine whether the approximation is accurate or how much approximation should be performed in the end.

OBJECT OF THE INVENTION The present invention solves the above-mentioned conventional problems by providing model generation means that generates a plurality of training models for one spoken word, and by providing model generation means that generates a plurality of training models for one spoken word. It is an object of the present invention to provide a speech training device in which the training model can be varied according to the needs of the speaker and the speaker himself/herself can easily proceed with the training.

Structure of the Invention The present invention provides a signal for determining the degree of agreement between the utterance feature data of a speaker and a training model. A training model generation unit that generates a training model, an utterance feature extraction unit that extracts utterance features of a speaker, and a comparison between the training model generated by the training model generation unit and the utterance feature data extracted by the utterance feature extraction unit. an utterance comprising: a comparing section; and a matching degree determining section that receives the comparison data of the comparing section as input, - determines the degree of matching within a predetermined time of the utterance, and sends the matching degree determination result to the training model generating section. It is a training device, and by changing the training model according to the progress of the speaker's training, the speaker can easily proceed with the training himself/herself.

DESCRIPTION OF THE EMBODIMENTS FIG. 1 is a block diagram showing the configuration of a speech training device according to an embodiment of the present invention. In FIG. 1, 1 is a training model generation unit that sequentially generates training models from low to high difficulty φIt for the basic training model of utterance, and 2 is an utterance feature that extracts the utterance features of the speaker. The extraction unit, which is omitted here, extracts utterances such as laryngeal vibrations and nasal vibrations from the detection signals of laryngeal vibration sensors, nasal vibration sensors, etc., which are attached to the vocal organs of each part of the speaker. This is used to extract features. 3 is a comparison unit that compares the training model generated by the training model generation unit 1 and the feature extraction data of the utterance feature extraction unit 2; 6 is a matching degree determining unit that determines the matching degree between the training model and the feature extraction data in one utterance based on the comparison result of the comparing unit 3 within 6 seconds); 6 is the training model, the utterance feature extraction data, and the matching degree; A display control unit 6 is a display device such as a CRT for displaying the determination result on a display device.

2 to 4 show an example of the display screen of the speech training device of this embodiment, and FIG. 2 is an example of the waveform display of the basic training model of the spoken language. It has been created. Figures 3(a) to (C
) is for the waveform display example of the basic training model in Figure 2.
The difficulty level as a training model is increased from a) to (C). That is, in FIG. 3(a), the starting point 31. of the utterance is the characteristic point of the waveform. End point 32
The goal of speech is set at five locations, such as 33° and 34.35 points of change, and the speaker's feature extraction data imitates the above goals at the same coordinates. .

FIG. 3(b) is a version with a higher level of difficulty than the one shown in FIG.

FIG. 3(C) most closely approximates the basic training model shown in FIG. 2, and is a level in which the training objective can be achieved if the utterance target at this level can be imitated to some extent.

Figures 4 (a) to (C) show (-) to (C) for the waveform display example of the basic training model in Figure 2, similar to Figure 3.
In this case, the difficulty level is expressed as a width W centered on the basic training model.

Regarding the speech training device of this embodiment configured as shown in FIG. 1, its operation will be described below as shown in the display example in FIG.
It is shown together with a flowchart showing the operation of the figure.

7 <- (a)... Although omitted in Figure 1,
This process is started by a training start process request.

(b) The training model generator 1 sends the first step training model with the lowest difficulty level as shown in FIG. 3(a) to the comparison unit 3 and the display control unit 6 will be sent to.

(c)...The display control unit 5 displays the training model on the display screen of the display device 6.

)...Display control unit 5 and uniform layer determination unit 4
is initialized, and the display control unit 6 sets the display start end point of the feature extraction data extracted by the utterance feature extraction unit 2 to the left end point of the screen, which is the starting point of the same coordinates as the training model. , - The scattering determination unit 4 sets the cumulative count value of the degree of coincidence to zero.

(E)...The display control section 6 inputs the feature extraction data of the speech feature extraction section 2 at a predetermined sampling period to improve the display image of the display device 6 (in this case, the third
Displayed at the same time 41+ of Japanese Patent Application Laid-Open No. 163078 (3). Further, this sampling timing signal is sent to the comparison section 3 and the uniform layer determination section 4.

(f)...The comparison unit 3 compares the target points (31 to 36) of the training model at that time with the utterance feature data using the -J- sampling timing signal, and compares the comparison results. It is sent to the uniform layer determination section 4.

(G) ...... - The scattering determination section 4 judges the comparison result, and if the training model target data (Dm) and the utterance feature data (Dr) match, - the scattering counting section (force branch,
If there is a mismatch, -Scattered layer display section (skip to 1 sword).

(g)...-The diffused layer determination unit 4 determines the above (Dm) and (
D, ) match, 1 is incremented in the -dispersed old church register.

(Male) The display control section 6 displays the value of the -dispersed layer count register on a part of the display screen of the display device 6 (FIG. 3 38).

C)...The display control unit 5 determines whether the predetermined frame 9 of the uttered word is finished, and if it is not finished, the display control unit 5 displays the above 0 Ukara (I
The process J) is repeated, and the feature data imitates the right direction (ie, the time axis) of the model being displayed in FIG. 3(a).

Q... - When a predetermined frame of the spoken word ends, the display control section 5 gives an end command to the uniform layer determination section 4,
- The scattered layer determination section 4 sends the value of the coincidence count register (a) counted so far to the training model generation section 1. The training model generation unit 1 compares the value of (6) above with the value of the request-dispersion in the training step.

(7)...The training model generation unit 1 makes a comparison judgment between the above (a) and proviso, and if (K)<(6), the training goal at that step has not been achieved. is determined and the display control unit 5 is notified of the determination. The display control unit 5 thereby repeats the process from (c) redisplaying the training model in the same step to (w).

(W)... (B) In the case of ≧ (R), the training model generation unit 1 determines the end of the training of step 1o \-, and creates a training model for the next step in the same utterance. Determine the presence or absence of.

(a))...If there is a training model for the next step, the training model generation section 1 sends the model to the comparison section 3 and display control section 5. The next step model in this case is the display example shown in Figure 3 (1).
The process proceeds sequentially to the steps in the display example shown in FIG. When the next step model is loaded, the display control unit 5 repeats the process from (c) again.

(→...If there is no model for the next step, move on to training processing for the next new spoken word.

The comparison and determination of the training model target and the utterance feature data will be described in detail with reference to the display example shown in FIG. 3 (-). Basic training waveform 30
Assuming that the speaker's utterances are made as shown by the dotted line 39 in the figure with respect to the training targets 31 to 36 set on the line -,
As can be seen from the figure, 4 out of 5 target points matched. In this case, 11 <-:' matching refers to when the feature data overlaps the target point (hatched circle) even once.

Figure 6 QO requirements - The scattering does not necessarily have to be 100%, for example, if the difficulty level is high, it may be so%.

As described above, according to this embodiment, by providing a training model generation unit that generates a plurality of training models for the same spoken word, advanced training can be performed step by step according to the progress of training. You can proceed to.

In this example, the shape of a diagonal circle O is used as the target of the training model, but this is further changed to the shape of grass, and the shape of a horse is used as the feature data of the speaker. It is also possible to provide entertainment like an animation game, such as eating grass. At this time, the coincidence degree count 38 in FIG. 3 does not necessarily need to be displayed.

Effects of the Invention The speech training device of the present invention selects a training model from a low difficulty training model to a high difficulty training model for a basic training model of speech, based on a signal for determining the degree of agreement between the speech feature data of the speaker and the training model. A training model generation unit that sequentially generates a training model, an utterance feature extraction unit that extracts utterance features of a speaker, and a comparison between the training model generated by the training model generation unit and the utterance feature data extracted by the utterance feature extraction unit. a comparison section to
By providing a match determination section that takes the comparison data as input, performs a match determination within a predetermined time period corresponding to the utterance, and sends the scattered layer determination results to the training model generation section, the training model can be trained from the initial stage. Since the training model changes according to the progress of training through sequential steps, speakers themselves can easily progress through the training from the initial stage to the final stage, which has a great practical effect. .

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of a speech training device according to an embodiment of the present invention, FIG. 2 is a diagram showing a display example of a basic speech training model, and FIGS. Figures 4a-C are diagrams showing other display examples of the training model similar to those in Figures 3a-C, It is a flowchart diagram of one embodiment. 1... Training model generation unit, 2... Speech feature extraction unit, 3... Comparison unit, 4...・・−
Diffused layer determination section, 5...Display control section, 6...
・Display device. Patent applicant: Director of the Agency of Industrial Science and Technology Kawa 1) Hirobe Figure 2 Prisoners 3-4 (a-) (b)

Claims (1)

[Claims] a training model generation unit that sequentially generates training models from less difficult to more difficult based on a basic training model of utterances based on a match degree determination signal between the speaker's utterance feature data and the training model; a utterance feature extraction unit that extracts the utterance features of the utterance feature; a comparison unit that compares the reduction model generated by the training model generator with the utterance feature data extracted by the utterance feature extraction unit; Taking the comparison data as input, - determining the degree of matching within a predetermined time period of the utterance;
A speech training device comprising: a matching degree determining section that sends the matching degree determining result to the training model generating section, and making the training model variable according to the progress of speech training.