JPH0357777B2 - - Google Patents

Description

[Detailed Description of the Invention] Industrial Application Field The present invention relates to a speech training device for speech-impaired people.
In particular, the present invention relates to a speech training device that uses speech information from other speech organs as well as sound waves.

Structure of the conventional example and its problems It is estimated that there are several hundred thousand people with language disabilities in Japan who are unable to speak due to hearing impairment, etc. However, due to the lack of language trainers and the long training period, etc. However, it is difficult to say that sufficient training time is secured. For this reason, it is desired to develop an effective speech training device, but a training device that can classify and recognize speech sounds and display whether the speech is correct or incorrect has not yet been obtained.

In recent years, as a device closely related to speech training devices,
There is a speech recognition device that is undergoing active research and development. Most of the ordinary speech recognition devices use only the information of the speech waves, but from the speech waves alone, the consonants /p, t, k/, /b, d, g/, /m, etc. n,
Accurate detection of η, etc. and mutual discrimination are difficult. For this reason, a pronunciation feature extraction device that uses information such as vocal cord vibration, nasal vibration, oral airflow, presence or absence of contact between the tongue and palate, etc. in addition to voice waves (see Japanese Patent Application Laid-Open No. 150997/1983)
etc. have been proposed.

FIG. 1 shows the criteria for phoneme classification in the conventional pronunciation feature extraction device. That is, in FIG. 1, voiced sounds /b, d, g, m,
n, η/ has vocal cord vibration (+), and is a voiceless sound/
For p, t, k, and h, there is no vocal cord vibration (-). Nasal sounds /m, n, η/ have nasal vibration (+), /
p, t, k, b, d, g, h/ have no nasal vibration (-). Plosive sounds/p, t, k, b, d,
For g/ and the fricative sound /h/, the flow velocity of the oral airflow is present (+), but for /m, n, and η/, it is absent (-). Furthermore, the rate of change in the oral airflow velocity is such that /p, t, b, d/ are large and present (+), but /k, g, h/ are small and are absent (-). For palatal contact, /p, b,
m, h/ are not closed patterns, /t, d,
In n/, there is a pattern of anterior tongue closure, and in /k, g,
η/ results in a pattern of posterior tongue closure. Therefore, it is assumed that phonemes can be classified according to FIG.

However, according to experiments, with the above configuration, (1) voiced plosives /b, d, g/ have a strong tendency to become nasalized, and nasal vibrations may be present (+).

(2) Posterior tongue closure at /k, g, η/ may not be detected at all due to individual differences, and in this case, no closure occurs.

(3) /k, g/ where the point of articulation of rupture is in the posterior lingual region,
The detection of the flow velocity of the oral airflow is somewhat unstable compared to /p, t, b/, and [flow velocity, rate of change] is [present (+), present (+)], [present (+), absent]. (-)], [No (-), No (-)], etc. may vary.
A similar tendency is also seen for /d/, although it is less frequent.

Due to these reasons, the criteria shown in Figure 1 is that /p/ and /k/, /d/ and /
It has been found that it is difficult to use it as is for speech training because confusion occurs between n/, /g/, /η/ and /m/.

Purpose of the Invention It is an object of the present invention to provide a speech training device that can classify and recognize speech sounds and display whether they are correct or incorrect, in accordance with actual speech training, by solving the difficulties in applying the conventional speech recognition device. do.

Composition of the Invention The present invention is a speech training device equipped with a detector for information on sound waves, nasal vibrations, laryngeal vibrations, and closed contact between the palate and the tongue, a classification recognition unit, and a training sound designation unit,
By performing classification recognition focusing only on phonemes that are phonetically similar to the designated training sound, difficulties in speech recognition can be eliminated and classification recognition that is suitable for actual training can be realized.

DESCRIPTION OF THE EMBODIMENT FIG. 2 is a block diagram showing an embodiment of the present invention. In Fig. 2, 1 is a sound wave detector to obtain information on the presence or absence of sound and its onset, 2 is a nasal vibration detector attached near the center of the nasal wall and used to obtain information on nasal sounds, and 3 is a laryngeal vocal cord. A throat vibration detector is installed nearby to obtain information on voiced and unvoiced sounds; 4 is a rupture detector for obtaining information on rupture by detecting, for example, the rate of change in flow velocity from an oral airflow detector located in front of the oral cavity; 5 is a tongue closure detector for obtaining closure contact information from a palate contact detector attached to the palate in the oral cavity in order to detect contact between the tongue and the palate; 6 is the above-mentioned 1 to 1;
5, a classification recognition unit that performs classification recognition of speech sounds from the detection information; 7, a speech correctness determination result display unit that shows the results of the classification recognition unit; 8, a training sound designation unit that accepts the designation of a training sound prior to training; 9; is a control unit that controls the classification recognition unit 6, the correct/incorrect determination result display unit 7, and the training sound designation unit 8.

The operation of the classification recognition unit 6 in the embodiment of the present invention configured as described above will be explained in detail below according to an example.

Speech training does not involve practicing various phonemes in a chaotic manner, but is carried out systematically, taking into account the similarities, contrasts, and difficulty of the phonemes. FIG. 3 shows an example of the Pabama table, which is the most basic and important training goal in speech training. Training begins with /p/ and moves to adjacent sounds in rows and columns as the user becomes proficient.

FIG. 3 has common features depending on the rows and columns. That is, from the point of articulation position, /p, b, m/ are bilabial sounds, /t, d, n/ are alveolar sounds, and /k, g, n/ are soft palatal sounds. In terms of types of sounds, /p, t,
k/ is a voiced plosive / b, d, g are voiced plosives, /
m, n, η/ are nasal sounds. As can be inferred from these points, even in actual speech training, confusion tends to occur between sounds that are adjacent in the row and column directions, and confusion between sounds that are far apart is rare. For example, when uttering /p/, it may be mistaken for /t/ or /b/, but it is extremely rare for it to be mistaken for /k/, /m/, or other sounds. In other words, according to the actual training/
When training p/, /p/ (correct answer), /t/ or b/ (mistake to adjacent sound), /? / (other mistakes) classification recognition, or /p/, /
t/, /b/, /? / performs classification recognition,
Sufficiently effective training can be performed without performing difficult classification recognition of /p/ and /k/.

Similarly, if you know the speech sounds to be trained, /
Training is possible without performing classification recognition of g/ and /m/, /η/ and /m/.

Next, /d, g/ have a strong tendency to become nasalized, and sometimes show nasal vibrations to the same extent as nasal sounds, causing errors; however, according to experiments, /d, g/ have a strong tendency to become nasalized. / It was found that strong nasal vibrations rarely lasted for more than 100 msec, and conversely, for nasal sounds, strong nasal vibrations lasted for more than 120 msec in most cases. Therefore, using this fact, /d/ and /
Confusion between n/, /g/ and /η/ can be almost eliminated.

Based on the above-mentioned concept, FIG. 4 shows the types of judgment results and the combinations of separation functions used for judgment when classification recognition is performed for the training of each sound shown in FIG. 3.

In the training for /t/, /p/ and /k/ are not classified and recognized, and /p and k/ are used as the judgment results. As mentioned above, this means that training can be carried out even if /p and k/ are added as errors to adjacent sounds, and from a training standpoint, it is a method that does not make incorrect correct/incorrect judgments even if the performance of classification recognition is reduced. This is because it is desirable and the classification recognition unit can be realized simply. /
The same is true for /b, g/ in the training of d/, and /m and η/ in the training of /n/.

Figure 5 illustrates the criteria for realizing each separation function in Figure 4. For example, the voiceless plosive /p/ has no nasal vibration, is plosive, is voiceless, and has a tongue This indicates that there is no closed contact. Tongue closed contact information is used as the separation condition between columns.
In other words, the criteria for separating /p, b, m/ from /t, d, n/ are the absence and presence of tongue closing contact, /t, d, n/
The criterion for separating /k, g, η/ is the presence or absence of tongue closing contact. /p, t, k/ and /b, d,
The conditions for determining the line spacing of g/ are unvoiced and voiced. The criteria for separating the lines between /b, d, g// and /m, n, η/ are a tendency to nasalization and strong and long nasal vibrations.

In addition, the separation function uses the characteristics that both have in common as a common basic condition, and if the common basic condition is not satisfied,
The classification recognition becomes "?" (other).

FIG. 6 summarizes common basic conditions and separation judgment conditions for each separation function based on FIG. 5.

The operation of the speech training device of this embodiment, which is equipped with the classification recognition section configured as above, will be explained below using a flowchart.

FIG. 7 shows an example of /p/ training. This will be explained using FIGS. 2 and 7.

(a) First, prior to training, the training sound specifying section 8 specifies which sound is to be trained. what if/
If p/ is specified, the classification recognition unit 6 is notified via the control unit 9 that training for /p/ will start.

(b) When the training sound is uttered next, the classification recognition unit 6 receives the information from the detectors 1 to 5 and starts classification recognition of /p/.

(c) In the classification recognition of /p/, first, the p/t separation function is activated, and the basic conditions and separation judgment conditions shown in FIG. 6 are checked. If the basic condition is violated, the violation condition is secured and the process proceeds to p/b separation 2.

Next, the separation judgment condition is checked, and if there is a tongue closed contact, it is judged as /t/, and the judgment sound /t/ and the reason for the judgment are displayed on the correctness judgment result display part 7 through the control unit 9. .

(d) In the p/b separation function, the conditions shown in Figure 6 can be examined. If you violate the basic conditions, keep the violation conditions and proceed to next step. In the separation judgment condition, it is checked whether or not there is a voice, and if it is voiced, it is judged as /b/, and the correct/incorrect judgment result display section 7 is displayed through the control section 9.
Along with the judgment sound /b/, the judgment reason "Yusei" is shown.

(E) It is checked whether or not there is a violation condition using the p/t separation function, and if there is a violation condition, the process proceeds to step (5). If there is no violation, the answer is correct. /p/ is displayed on the correct/incorrect determination result display section 7.

(f) This processing step violates the basic conditions. The correct/incorrect judgment result display section 7 shows /? / indicates the content of the violation. The separation functions that are activated in the training of /p/ are the p/t separation function and the p/b separation function, and violations of the basic conditions of these separation functions are “pion”, “haretsei nashi”,
There are four types: “Yusei” and “Shitaheisa Ali”.

FIG. 8 shows the types of messages that violate the basic conditions for each training sound.

Effects of the Invention The present invention includes a classification recognition unit that classifies speech sounds using the outputs of each detector as input, and in this classification recognition unit, a specific separation function is set according to the training sound designated by the training sound designation unit. Since the configuration is configured to select and classify speech sounds that are difficult to recognize, it is easier to classify and recognize speech sounds that are difficult to recognize. It is possible to provide a high-performance speech training device at a low cost, and its practical value is high.

[Brief explanation of drawings]

Fig. 1 is a diagram showing the criteria for phoneme classification in a conventional pronunciation feature extraction device, Fig. 2 is a block diagram showing an utterance training device in an embodiment of the present invention, and Fig. 3 is a diagram of the Pabama table. An array diagram showing an example, FIG. 4 is a correspondence diagram showing the types of judgment results for each training sound and the types of separation functions used to obtain the judgments in the embodiment of the present invention, and FIG. FIG. 6 is a diagram illustrating the common basic conditions and separation determination conditions based on FIG. 5. FIG. 7 is a flowchart showing an example of the operation of this embodiment. The figure is a correspondence diagram showing the types of violation messages for each training sound. 1... Sound wave detector, 2... Nasal vibration detector, 3
...Laryngeal vibration detector, 4... Rupture detector, 5...
. . . Tongue closure detector, 6 . . . Classification recognition unit, 7 .

Claims (1)

[Claims] 1. A sound wave detector, a nasal vibration detector, a laryngeal vibration detector, a rupture detector, and a tongue closure detection that detects closure contact information between the tongue and the hard palate. a classification recognition unit that has a plurality of separation functions depending on the training sounds and classifies the speech sounds using the outputs of the respective detectors as input; a training sound specification unit that designates the training sounds; and a correct/incorrect judgment result display unit that shows the result, and the classification recognition unit selects a specific separation function according to the training sound designated by the training sound designation unit to perform classification. Device.