JPS5946699A - Word voice registration system - 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] (1) Technical field of the invention The book 'jtiEiJ is a single 6! The present invention relates to a word speech registration method that comprehensively improves the recognition rate by taking into account that the recognition rate of words varies depending on the difference in the number of syllables during registration.

(2) Prior art and problems Conventional word speech recognition devices It can register each word once, and register all single words twice.
The tjfj method has the disadvantage of not achieving a sufficient recognition rate, especially for words with few phonetic patterns, and the latter improves this point, but has the disadvantage of requiring a large number of registrations and being too time-consuming. Ta.

(3) Purpose of the Invention The purpose of the present invention is to change the number of syllables when registering words.
It is an object of the present invention to provide a word voice registration method that comprehensively improves the recognition rate considering the difference in the recognition rate of I.

(4) Structure of the Invention In order to achieve the above-mentioned object, the word voice registration method of the present invention registers condolence candy with a small number of syllables in a single mri tf voice recognition device that registers and recognizes a single ml voice. The present invention is characterized in that it is provided with means for creating a plurality of dictionary contents or the average dictionary contents by repeating the process.

(5) Examples of the invention Figure FA1 and Figure 2 are diagrams celebrating the principle of the present invention.

Figure 1 is a diagram showing the relationship between the syllables of the word F and the recognition rate. That is, in general, when the number of sounds Nj)Iil in a word speech is small, the recognition rate is low. On the other hand, if we prepare a dictionary Uchitani whose registration frequency is 1, 2, and 3 times, the recognition rate improves as shown in @, ■, and ■ in the same figure.

Furthermore, for word sounds with a large number of syllables, the recognition rate does not improve even if the number of registrations is increased.

Therefore, only words with a small number of syllables may be registered four times in multiple countries, and words with a large number of syllables may be registered only once.

FIG. 2 shows an example of this method. That is,
1-syllable and 2-syllable words can be registered 6 times; 6-syllable and 4-syllable words can be registered 2 times; 25-syllable or more words can be registered once without increasing the number of registrations for all words. A high recognition rate can be maintained.

FIG. 6 is an explanatory diagram of the configuration of an embodiment of the present invention.

In the figure, a word name code and the number of syllables of the word are stored in each address of the word name/syllable count list memory 10, and the contents specified by the word name syllable count list memory address counter 20 of the r-speed transmission are , registration instruction word register 1
1 and syllable number register 12, respectively. The contents of the registration instruction word register 11 are displayed on the display device 1.
5 is displayed. From the value of the syllable number register 12, the number of registrations of the word is determined using the syllable number registration number conversion ROM 14, and the value is stored in the registration number register 15. The table below shows an example of the contents of the syllable registration number conversion ROM 14.

On the other hand, the voice V uttered based on the registered command word displayed on the display device 16 is sent by the microphone 16 to the voice analysis device 1t17 for analysis, and 11!
It is converted into an M-pattern by the J-character extraction device 18. Resignation V
) Pattern 7 lJ: Registration instruction ifl# Stored together with the contents of the register 11 in the memory μ memory 22 specified by the memory address counter 21. Diction-J memory address color/re 21t:J: is incremented by the voice registration pulse, and the registration number counter 15 is decremented by the voice registration pulse. The next zero detection unit 19 outputs a nox when the content of the registered number of times 15 becomes zero,
'4' Increment the ttM name syllable list memory address counter 20 and set the registration instruction word to the next word. Word registration is performed by repeating the above operations.

According to this embodiment, since the ROM is used for changing the number of registered syllables, the correspondence between the number of syllables and the number of registrations can be determined without changing the word name code list. It has the advantage of being able to be changed freely by simply changing the settings.

In the embodiment, the registration of words with a small number of sound rtiTs is performed multiple times, but by calculating the average of the contents of these multiple registered words and storing it in the dictionary, the same effect can be achieved with less retrieval capacity. be able to.

(6) Effects of the Invention As explained above, according to the present invention, the single jf7”+
By increasing the total number of registrations for words with a small number of sounds during voice registration, it is possible to maintain a high recognition rate with a relatively small number of registrations overall.

[Brief explanation of drawings]

Figures 1 and 2 are explanatory diagrams of the configuration of an embodiment of the present invention. In the figures, 10 is word name syllable number list memory, and 11 is registration. Instructions 1f 74 registers, 12 back 4 registers, 13e display device,
14 is the sound 'KJ number registration frequency conversion ROIVI, 15 is the registration frequency counter, 16t is the microphone, 17&J,
Speech analysis device, 18//i, feature extraction device + ffj, 'I
s) is a zero detection nL; 20 is a word name syllable number list memory address counter; 21 is a memory address counter for μs;

Claims (1)

[Claims] In a word speech recognition device that performs recognition by registering word sounds, it is a percent feature that a means is provided to register words with a small number of syllables multiple times and create the dictionary contents of selected fields or the average dictionary contents. Word audio registration method.