JPS6162166A - Japanese language sentence writing device - Google Patents

Abstract

PURPOSE:To obtain a Japanese sentence writing device that enables retrieving of an independent word dictionary at high speed by making identification of independent word vowels by using an independent word vowel string dictionary that stores in correspondence to a vowel string expressed by only vowel parts. CONSTITUTION:When 'setsubun' is inputted to a monosyllable recognizing section 1, it is supported that candidates of recognition are 'se', 'e', 'he', 're' for the first inputted syllable 'se'. These candidate group are stored in a monosyllable buffer 2. An independent word candidate preparing section 3 prepares independent word candidates from combination of the monosyllable recognition candidates. On the other hand, a vowel 'e' is extracted from the first inputted syllable 'se' by a vowel extracting section 7, and a vowel string 'euun' is stored in a vowel buffer 8 by similar extraction. Then, an independent word vowel identifying section 10 retrieves an independent word vowel string dictionary 9 for the vowel string 'euun'. In the case of this example, the vowel sting 'euun' in the dictionary, and its corresponding character string is only one, that is, an independent word 'setsubun'.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a Japanese sentence creation device that enables high-speed correction of speech recognition errors when a monosyllable is input.

Conventional configurations and their problems In recent years, voice input has attracted attention from the perspective of man-machine interfaces, and the creation of sentences by voice has been researched and developed. “A conventional Japanese text creation device will be explained below.

FIG. 1 shows a block diagram of a conventional Japanese sentence creation device, in which 1 is a monosyllable recognition unit that recognizes input monosyllables. A monosyllable candidate buffer 2 stores recognition candidates obtained by the monosyllable recognition unit 1. Reference numeral 3 denotes an independent word candidate creation unit that creates independent word candidates from combinations of recognition candidates stored in the monosyllable candidate buffer 72. 4 is an independent word dictionary that stores a plurality of independent words including kanji in correspondence with character strings representing their pronunciations. Reference numeral 5 denotes an independent word search unit that searches the independent word dictionary 4 with the independent word candidates created by the independent word candidate creation unit 3 to obtain independent words containing the corresponding Chinese characters. Reference numeral 6 denotes a sentence creation unit that creates sentences using the independent words including kanji obtained by the independent word search unit 5.

The operation of the conventional Japanese text creation device configured as described above will be explained below.

Let us consider a case where /seppun/ is input to the monosyllable recognition unit 1. Assume that the recognition candidates for the first human syllable ``se'' are ``se'', ``e'', ``he'', and ``re''. Second human syllable “
Suppose that the recognition candidates for ``tsu'' are ``su'', ``u'', ``tsu'', and ``fu''. Similarly, it is assumed that a recognition candidate group as shown in Table 1 is obtained. These candidates are stored in a monosyllabic candidate buffer.

Table 1 The independent word candidate creation unit 3 creates independent word candidates from the combination of these monosyllabic recognition candidates. One of these creation methods
One is disclosed, for example, in Japanese Patent Application No. 59-36447. The independent word candidates obtained by the independent word candidate creation section 3 are shown in Table 2.

Table 2 Monosyllabic recognition candidates if all combinations are considered
・If the number is 5 m and the input syllable is n, then M = m'' candidates will be obtained. In the case of this conventional example.

M=4’=256 Candidate It will exist.

The independent word search unit 5 reads out the candidates in Table 2 one by one,
Search the independent word dictionary 4. If it does not exist, it will be rejected,
Sequentially read and search candidates.

However, the above configuration has the following problems.

(1) There were problems with processing speed due to the large number of searches in the independent word dictionary. If the value of the maximum number of searches M is reduced in order to improve speed, the probability that a correct independent word cannot be found from recognition candidates increases.

(2) If the correct uttered syllable is not included in the monosyllable candidates sent from the monosyllable recognition unit 1, the correct independent syllable is selected among all the independent word candidates obtained by the independent word candidate creation unit 3. Since the word does not exist, M unnecessary independent word searches are performed.

Because of the above problems, it has been desired to reduce the number of independent word dictionary searches, and it has also been desired to be able to make correct corrections even when the correct uttered syllable is not included in the monosyllable candidates.

OBJECT OF THE INVENTION The present invention solves the above-mentioned conventional problems.It is an object of the present invention to speed up search processing of an independent word dictionary by verifying independent word vowels using a vowel string and an independent word vowel string dictionary. The purpose is to provide a Japanese sentence creation device that enables.

Structure of the Invention The present invention provides an independent word vowel string dictionary that stores character strings representing the readings of independent words in correspondence with vowel strings represented only by vowel parts when the readings are uttered, and acoustic processing. a monosyllable recognition unit that recognizes input monosyllables by performing the following steps; a monosyllable candidate buffer that sequentially stores the recognition candidates obtained by the monosyllable recognition unit; and a monosyllable candidate buffer that sequentially stores the recognition candidates obtained by the monosyllable recognition unit; an independent word candidate creation unit that creates independent word candidates by combining;
a vowel extraction unit that extracts the vowel part of the input monosyllable; a vowel buffer that sequentially stores the vowels extracted by the vowel extraction unit; and the independent word vowel string dictionary using the vowel strings stored in the vowel string A Japanese sentence creation device comprising an independent word vowel verification unit that searches for the presence or absence of the vowel string or a character string representing the pronunciation of an independent word corresponding to the vowel string and the reason for the independent word, By verifying independent word vowels using a vowel string and an independent embryonic sound string dictionary, independent word retrieval is possible even if an independent word can be uniquely determined from a vowel string even if there is a recognition error in the consonant part. The number of times becomes 1,
Since there is no need to perform a wasteful independent word search for vowel strings that do not exist due to vowel recognition errors, processing speed can be increased.

DESCRIPTION OF EMBODIMENTS FIG. 2 shows a block diagram of a Japanese text creation device according to an embodiment of the present invention. In FIG. 2, 7 is a vowel extraction step that obtains a vowel by extracting a vowel from the recognition data obtained by the monosyllabic recognition unit 1 or by converting the monosyllabic candidate characters stored in the monosyllabic candidate buffer 2 into vowels. Department. 8
is a vowel buffer that sequentially stores vowels extracted by the vowel extractor. Reference numeral 9 is an independent embryonic sound string dictionary that stores character strings representing the readings of independent words in correspondence with vowel strings expressed only by the vowel parts when the readings are uttered. 10 is a vowel string stored in the vowel buffer 8 and the independent word vowel verification unit 9
The independent word vowel verification unit searches for the presence or absence of the vowel string or obtains the character string representing the pronunciation of the independent word corresponding to the vowel string and the number of independent words.

The operation of the Japanese text creation device of this embodiment configured as described above will be described below.

For comparison with the conventional example, consider a first example in which the input speech and recognition results are the same as in the conventional example. In other words, “
Suppose that when inputting ``Seppun'', recognition results as shown in Table 1 are obtained, and from the combination of monosyllable recognition candidates in Table 1, independent word candidates as shown in Table 2 are obtained.

On the other hand, the vowel extraction unit 7 extracts the vowel "e" from the first labor syllable "se", and similarly extracts the vowel "u" from the second labor syllable "tsu" and extracts the vowel "u" from the second labor syllable "tsu", and extracts the vowel "u" from the second labor syllable "tsu", and extracts the vowel "u" from the second labor syllable "tsu", The vowel "u" is extracted from the syllable, and the vowel "n" is extracted from the fourth labor syllable "n." Therefore, the vowel string ``Euhun'' is stored in the vowel Tsuku Nofa 8.

Next, the independent word vowel verification unit 1o searches the independent embryonic sound string dictionary 9 for the vowel string "Eutsun". Then check the following:

■ Check whether the vowel string "Euhun" exists in the independent word vowel verification section.

■ If so, find out how many character strings exist that represent the pronunciation of the independent word corresponding to that vowel string.

■ If only one string exists, find out what that string is.

In this example, the vowel string "euhun" exists in the dictionary,
It turns out that there is only one corresponding character string, and that is the independent word ``seppun''.

Therefore, the independent word search unit does not need to read out the independent word candidates in Table 2 one by one and search the independent word dictionary 4, but instead searches the independent word dictionary using the character string "Seppu" determined by the independent word vowel verification unit. By searching 4, one independent word search for ``Setsubun'' can be obtained.

In this example, conventionally, it would take up to 256 independent word searches to obtain "Setsubun," but by verifying the independent word vowels, "Setsubun" can be obtained with one independent word search. It turns out.

A second example will be described below.

Table 3 If the recognition results shown in Table 3 are obtained when the human voice is ``Seppun'' as in the first example, there is no correct pronunciation among the output syllable candidates for the second human labor syllable ``tsu''. Since no syllables are included, the correct independent word "seppun" does not exist among all the independent word candidates obtained by the independent word candidate creation unit 3. Therefore, with the conventional method, even if independent word searches are performed 256 times, the independent word "Setsubun" cannot be obtained. On the other hand, by extracting the vowels of the input monosyllables and verifying the independent word vowels, it is possible to obtain the independent word "seppun" as in the first embodiment. ” can be obtained.

A third example will be described below.

Table 4 Suppose that when the human voice is ``takageta'', the recognition results shown in Table 4 are obtained, and from the combinations of monosyllabic candidates in Table 4, the independent word combinations shown in Table 5 are obtained. .

Margin Table 5 In short, this example is an example in which a vowel recognition error occurred as the vowel of the first human labor syllable as "o" instead of "a". On the other hand, the vowel extraction unit 7 extracts the vowel of the input syllable,
The vowel string "oaeaa" is stored in the vowel buffer 8.

Next, the independent word vowel verification unit 10 searches the independent embryonic sound sequence dictionary 9 for the vowel string “oaea” □ As a result,
It can be seen that the vowel string ``oaeaa'' does not exist in the dictionary. Conventionally, it would be necessary to perform an independent word search using 266 independent word candidates as shown in Table 6 to find out that no corresponding independent word exists. This eliminates the need for word dictionary searches.

In addition, in the above three examples, the reduction in the number of searches when searching an independent word dictionary was described, but the need for searching is not limited to the independent word dictionary, and it is also possible to verify vowels using a vowel string dictionary. Anything that can be done is fine. For example, the number of searches can be reduced by using the attached embryotoxin dictionary when searching the attached word dictionary and verifying vowels.

Effects of the Invention As is clear from the above explanation, the Japanese sentence creation device of the present invention is equipped with an independent embryonic sound sequence dictionary and verifies independent word vowels, thereby reducing the number of searches of the independent word dictionary. Processing can be made faster, and its practical effects are great.

[Brief explanation of the drawing]

Figure 1 is a block diagram of a conventional Japanese text creation device.
FIG. 2 is a block diagram of the Japanese text creation device of the present invention. 1... Monosyllabic recognition unit, 2... Monosyllabic candidate buffer, 3... Independent word candidate creation unit, 4...
...Independent word dictionary, 5...Independent word search section, 6
......Sentence creation section, 7...Vowel extraction section,
8...Vowel buffer, 9...Independent embryotoxin sound sequence dictionary, 1o...Independent word vowel verification section. Name of agent: Patent attorney Toshio Nakao and 1 other person 1st
Figure 2

Claims (1)

[Scope of Claims] An independent word vowel string dictionary that stores a character string representing the reading of an independent word in correspondence with a vowel string expressed only by the vowel part when the reading is uttered, and an acoustic processing. a monosyllable recognition unit that recognizes an input monosyllable by applying a single syllable recognition unit; a monosyllable candidate buffer that sequentially stores recognition candidates obtained by the monosyllable recognition unit; and a combination of the monosyllable candidates stored in the monosyllable candidate buffer. an independent word candidate creation unit that creates an independent word candidate using the input monosyllable; a vowel extraction unit that extracts the vowel part of the input monosyllable; a vowel buffer that sequentially stores the vowels extracted by the vowel extraction unit; and an independent word vowel verification unit that searches the independent word vowel string dictionary using the stored vowel string to obtain a character string representing the presence or absence of the vowel string or the pronunciation of the independent word corresponding to the vowel string. A Japanese sentence creation device featuring: