JP7615841B2 - Voice recognition program and voice recognition device - Google Patents

Description

The present invention relates to a voice recognition program and a voice recognition device.

Patent Document 1 discloses a technology for performing speech recognition by providing a recognition dictionary for each type of word (address, facility name, telephone number, etc.) and searching input speech in the recognition dictionary corresponding to the type of word selected in advance by the user. Specifically, the recognition dictionary stores character strings (hereinafter referred to as "pronunciation character strings") that represent pronunciations using kana characters, alphabets, etc., and phrases that correspond to the pronunciation character strings. The corresponding phrase is obtained by searching the input string, which is a character string converted from the input speech, in the pronunciation character strings in the recognition dictionary. By providing a recognition dictionary for each type of vocabulary, the number of pronunciation character strings and phrases can be reduced compared to storing the pronunciation character strings and phrases of all types of vocabulary in a single recognition dictionary, and the time required for search can be shortened.

JP 2006-162782 A (for example, paragraphs 0012-0015, 0033-0037, Figures 1, 4, 5)

However, while the recognition dictionary is organized by word type, it also stores various pronunciation strings corresponding to the word. For example, for an input string of five characters, the recognition dictionary stores pronunciation strings of various numbers of characters, such as two or three characters, depending on the word. Therefore, in addition to referencing pronunciation strings with the same number of characters as the input string, a search must also refer to pronunciation strings with a different number of characters than the input string. This causes problems in that searches take time and there is a risk of a slower voice recognition response.

The present invention has been made to solve the above-mentioned problems, and aims to provide a voice recognition program and a voice recognition device that can improve the response of voice recognition.

In order to achieve this object, a speech recognition program of the present invention is a program for causing a computer having a storage unit to execute a speech recognition process, the storage unit being configured to function as a conversion dictionary storage means for storing a plurality of conversion dictionary data, which are dictionary data consisting of a combination of a pronunciation string and a word corresponding to the pronunciation string, the plurality of conversion dictionary data being divided by feature amount of the pronunciation string, and the program includes a speech conversion step of converting an input speech into an input string representing the pronunciation, a feature amount acquisition step of acquiring the feature amount of the input string converted in the speech conversion step, a dictionary acquisition step of acquiring from the conversion dictionary storage means conversion dictionary data corresponding to the feature amount acquired in the feature amount acquisition step, and ... data acquired in the dictionary acquisition step the conversion dictionary data corresponding to the input string converted in the speech conversion step. The conversion dictionary data is created from all dictionary data in which combinations of all pronunciation strings included in a plurality of conversion dictionary data and words corresponding to those pronunciation strings are stored, and when the all dictionary data is updated, combinations of pronunciation strings and words corresponding to those pronunciation strings included in the all dictionary data after the update are rearranged based on features of the pronunciation strings, and conversion dictionary data for each feature of the pronunciation strings is created from the rearranged combinations of pronunciation strings and words corresponding to those pronunciation strings .

The speech recognition device of the present invention further comprises: a conversion dictionary storage means for storing a plurality of conversion dictionary data consisting of a combination of a pronunciation character string and a word or phrase corresponding to the pronunciation character string, the conversion dictionary data being divided into a plurality of conversion dictionary data for each feature of the pronunciation character string; a speech input means for inputting speech; a speech conversion means for converting the speech input by the speech input means into an input string representing the pronunciation; a feature acquisition means for acquiring a feature of the input string converted by the speech conversion means; a dictionary acquisition means for acquiring conversion dictionary data corresponding to the feature acquired by the feature acquisition means from among the conversion dictionary data stored in the conversion dictionary storage means; and a search means for searching for a word or phrase corresponding to the input string converted by the speech conversion means from the conversion dictionary data acquired by the dictionary acquisition means. and an output means for outputting the searched word if the search means is able to search for a word, and for outputting the input string converted by the voice conversion means if the search means is unable to search for a word . The conversion dictionary data is created from all dictionary data in which combinations of all pronunciation strings contained in a plurality of the conversion dictionary data and words corresponding to those pronunciation strings are stored, and when the all dictionary data is updated, the combinations of pronunciation strings and words corresponding to those pronunciation strings contained in the updated all dictionary data are rearranged based on the features of the pronunciation strings, and conversion dictionary data for each feature of the pronunciation strings is created from the rearranged combinations of pronunciation strings and words corresponding to those pronunciation strings .

According to the speech recognition program of claim 1, a plurality of conversion dictionary data are stored, which are dictionary data consisting of a combination of a pronunciation string and a phrase corresponding to the pronunciation string, and which are divided according to the feature of the pronunciation string. Then, the feature is acquired from the input string representing the pronunciation of the input voice, and conversion dictionary data corresponding to the acquired feature is acquired. Then, a phrase corresponding to the input string is searched for from the acquired conversion dictionary data, and if the phrase is found, the phrase is output, and if the phrase is not found, the input string is output.

That is, the conversion dictionary data used for searching for words and phrases stores pronunciation character strings of features common to the input string and the words and phrases, so that the conversion dictionary data stores only pronunciation character strings similar to the input string. This allows the conversion dictionary data to be quickly searched for words and phrases with pronunciation character strings that match the input string, thereby improving the response of speech recognition.
In addition, the conversion dictionary data is created from all dictionary data in which combinations of all pronunciation strings included in multiple conversion dictionary data and words corresponding to those pronunciation strings are stored, and when the all dictionary data is updated, the combinations of pronunciation strings and words corresponding to those pronunciation strings included in the updated all dictionary data are rearranged based on the features of the pronunciation strings, and conversion dictionary data for each feature of the pronunciation string is created from the rearranged combinations of pronunciation strings and words corresponding to those pronunciation strings.
The conversion dictionary data created in this way has the effect of being able to appropriately include combinations of pronunciation character strings updated in the entire dictionary data and phrases corresponding to the pronunciation character strings, and combinations of pronunciation character strings of features corresponding to the conversion dictionary data and phrases corresponding to the pronunciation character strings. In addition, since the conversion dictionary data is automatically created by updating the entire dictionary data, it is also possible to reduce the user's effort in updating the conversion dictionary data.

According to the speech recognition program of claim 2, in addition to the effect of the speech recognition program of claim 1, the following effect is achieved. An input speech is converted into an input character string broken down into words, and the feature quantities of each of the input character strings broken down into words are acquired. Then, conversion dictionary data corresponding to each of the feature quantities of the acquired input character strings broken down into words is acquired, and a phrase corresponding to each of the input character strings broken down into words is searched for from each of the acquired conversion dictionary data.

In other words, the features of the input string broken down into words are acquired, and phrases are searched for in the conversion dictionary data that corresponds to each of the acquired features. This has the effect of enabling more detailed and accurate phrase output compared to searching the conversion dictionary data directly for an input string containing multiple words.

According to the speech recognition program of claim 3, in addition to the effects of the speech recognition program of claim 1 or 2, the following effects are achieved. The feature value is the number of characters in the pronunciation string or the input string. As a result, the conversion dictionary data is stored according to the number of characters in the pronunciation string. Then, the number of characters in the input string is obtained, and the conversion dictionary data corresponding to the obtained number of characters is obtained and used for searching for words and phrases.

This makes it possible to omit searches for words using conversion dictionary data for pronunciation strings that do not match the number of characters in the input string, thereby enabling quick searches for words that match the input string. In addition, since the number of characters in the input string or pronunciation string can be easily obtained without complex analysis, conversion dictionary data that matches the feature value (number of characters) of the input string can be quickly obtained. These have the effect of improving the response of voice recognition.

According to the speech recognition program of claim 4, in addition to the effects of the speech recognition program of claim 1 or 2, the following effects are achieved. The feature is set to the first character of the pronunciation string or the input string. As a result, the conversion dictionary data is stored according to the first character of the pronunciation string. Then, the first character of the input string is obtained, and the conversion dictionary data corresponding to the obtained first character is obtained and used for searching for words and phrases.

This makes it possible to omit searching for words using conversion dictionary data for pronunciation strings that do not match the first character of the input string, thereby enabling quick searches for words that match the input string. In addition, since the first character of an input string or pronunciation string can be easily obtained without complex analysis, conversion dictionary data that matches the feature (first character) of the input string can be quickly obtained. These have the effect of improving the response of voice recognition.

According to the speech recognition program of claim 5, in addition to the effects of the speech recognition program of claim 1 or 2, the following effects are achieved. The feature quantity is a combination of the number of characters and the first character of the pronunciation string or the input string. As a result, the conversion dictionary data is stored according to the combination of the number of characters and the first character of the pronunciation string. Then, the combination of the number of characters and the first character of the input string is obtained, and the conversion dictionary data corresponding to the obtained combination of the number of characters and the first character is obtained and used for searching for words and phrases.

This makes it possible to omit searching for words in the conversion dictionary data for pronunciation strings that do not match the number of characters and first character of the input string, thereby enabling quick searches for words that match the input string. Furthermore, by using a combination of the number of characters and the first character as the feature, it is possible to reduce the number of pronunciation strings and their words stored in the conversion dictionary data compared to when the feature is only the number of characters or only the first character. This also enables quick searches for words that match the input string.

In addition, the number of characters and the first character of the input string or pronunciation string can be easily obtained without complex analysis, so conversion dictionary data corresponding to the features of the input string (combination of number of characters and first character) can be quickly obtained. This has the effect of improving the response of voice recognition.

The voice recognition device described in claim 6 achieves the same effect as the voice recognition program described in claim 1.

FIG. 2 is an external view of a mobile terminal. 1A is a diagram showing a schematic representation of the creation of conversion dictionary data from all dictionary data, and FIG. 1B is a diagram showing a schematic representation of the replacement of an input character string using the conversion dictionary data. FIG. 2 is a block diagram showing the electrical configuration of the mobile terminal. 13A is a flowchart of speech processing, and FIG. 13B is a flowchart of conversion dictionary data creation processing. 13 is a flowchart of a dictionary application process. 13A is a diagram showing a schematic representation of the creation of conversion dictionary data from all dictionary data in the second embodiment, and FIG. 13B is a diagram showing a schematic representation of the replacement of an input character string using conversion dictionary data in the second embodiment. FIG. 11 is a block diagram showing the electrical configuration of a mobile terminal according to a second embodiment. 13A is a flowchart of a conversion dictionary data creation process according to a second embodiment, and FIG. 13B is a flowchart of a dictionary application process according to the second embodiment.

A preferred embodiment of the present invention will now be described with reference to the accompanying drawings. First, the configuration of a mobile terminal 1 in this embodiment will be described with reference to FIG. 1. FIG. 1 is an external view of the mobile terminal 1. The mobile terminal 1 is an information processing device (computer) that performs voice recognition of speech uttered by a user H. The mobile terminal 1 is configured to be able to input a voice V, and converts the input voice V into an input string Ti, which is a string of characters representing the pronunciation. The converted input string Ti is replaced with an appropriate word or phrase W and displayed on the LCD 16 (see FIG. 3) by referring to conversion dictionary data Sd, which will be described later, in FIG. 2.

Next, the conversion dictionary data Sd and the replacement of the input character string Ti using the conversion dictionary data Sd will be described with reference to FIG. 2. FIG. 2(a) is a diagram that shows a schematic representation of the creation of the conversion dictionary data Sd from the entire dictionary data Ad, and FIG. 2(b) is a diagram that shows a schematic representation of the replacement of the input character string Ti using the conversion dictionary data Sd.

The mobile terminal 1 is provided with two types of dictionary data: full dictionary data Ad and conversion dictionary data Sd. The full dictionary data Ad stores a plurality of combinations of pronunciation strings Tp, which are strings of phonological characters such as hiragana and katakana, and words and phrases W, which are strings of kanji characters, alphabet characters, etc., that correspond to the pronunciation strings Tp. The full dictionary data Ad stores word-by-word words W and pronunciation strings Tp that correspond to the words and phrases W, and stores all combinations of words W and pronunciation strings Tp that are used to replace the input string Ti.

The conversion dictionary data Sd is dictionary data created from the entire dictionary data Ad, and is constructed by dividing the combinations of the pronunciation string Tp and the words and phrases W corresponding to the pronunciation string Tp (hereinafter abbreviated as "combinations of pronunciation strings Tp and words and phrases W") stored in the entire dictionary data Ad into groups based on the number of characters in the pronunciation string Tp.

That is, the conversion dictionary data Sd includes one-character dictionary data Sd1 that stores a combination of a pronunciation string Tp consisting of one character and a word or phrase W that corresponds to the pronunciation string Tp, and two-character dictionary data Sd2 that stores a combination of a pronunciation string Tp consisting of two characters and a word or phrase W that corresponds to the pronunciation string Tp. Similarly, three- to ten-character dictionary data Sd3 to Sd10 that store combinations of a pronunciation string Tp consisting of three to ten characters and a word or phrase W that corresponds to the pronunciation string Tp are also provided. Hereinafter, the one- to ten-character dictionary data Sd3 to Sd10 are collectively referred to as "M-character dictionary data SdM."

The input character string Ti is replaced using the conversion dictionary data Sd configured in this way. Specifically, as shown in FIG. 2(b), the input character string Ti is first decomposed into word-based input character strings TiN (N is a natural number) by morphological analysis. In the example of FIG. 2(b), the input character string Ti is "honjitsu arigatou gozaimasu." Therefore, by morphological analysis of this input character string Ti, the input character string Ti is decomposed into four words: input character string Ti1 consisting of "honjitsu," input character string Ti2 consisting of "wa," input character string Ti3 consisting of "arigatou," and input character string Ti4 consisting of "gozaimasu."

Then, by referencing each of the input character strings TiN broken down into words in the conversion dictionary data Sd, the corresponding phrase W is obtained. Specifically, the number of characters is obtained for each input character string TiN, and conversion dictionary data Sd is obtained that has a pronunciation character string Tp whose number of characters matches the obtained number of characters. Then, a search is performed between the input character string TiN and the pronunciation character string Tp in the obtained conversion dictionary data Sd, and the phrase W corresponding to the pronunciation character string Tp that matches the input character string TiN is obtained.

In the example of FIG. 2(b), the number of characters in the input string Ti1 is "4", so the four-character dictionary data Sd4 is obtained from the conversion dictionary data Sd, and the words and phrases W of the pronunciation string Tp that match the input string Ti1 are obtained from the four-character dictionary data Sd4. The number of characters in the input string Ti2 is "1", so the one-character dictionary data Sd1 is obtained, and the words and phrases W of the pronunciation string Tp that match the input string Ti2 are obtained from the one-character dictionary data Sd1. Similarly, the number of characters in the input strings Ti3 and Ti4 is "5", so the five-character dictionary data Sd5 is obtained, and the words and phrases W of the pronunciation string Tp that match the input string Ti3 and the input string Ti4 are obtained from the five-character dictionary data Sd5.

In this way, M-character dictionary data SdM is provided for each number of characters in the pronunciation string Tp, and each word or phrase W is searched for in the M-character dictionary data SdM that matches the number of characters in the input string TiN. By searching for a word or phrase W in the M-character dictionary data SdM that matches the number of characters in the input string TiN, it is possible to omit a comparison with the pronunciation string Tp, which has a number of characters that does not match the number of characters in the input string TiN, and therefore it is possible to quickly search for the word or phrase W.

In addition, since the number of characters in the pronunciation string Tp and the input string TiN can be easily obtained without complex analysis, it is possible to quickly create the conversion dictionary data Sd and quickly obtain the M-character dictionary data SdM that matches the number of characters in the input string TiN. This allows for a good response from when the user H inputs the voice V to when the word W corresponding to that voice V is displayed.

Furthermore, by decomposing the input character string Ti from the user H into word-by-word input character strings TiN and searching the conversion dictionary data Sd for each input character string TiN, it becomes possible to obtain a word-by-word term W that is more detailed and accurate than when an input character string Ti containing multiple words is searched for in the conversion dictionary data Sd.

Next, the electrical configuration of the mobile terminal 1 will be described with reference to FIG. 3. FIG. 3 is a block diagram showing the electrical configuration of the mobile terminal 1. As shown in FIG. 3, the mobile terminal 1 has a CPU 10, a flash ROM 11, and a RAM 12, which are each connected to an input/output port 14 via a bus line 13. The input/output port 14 is further connected to a microphone 15 for inputting voice V, an LCD 16 for displaying words and phrases W and the like resulting from voice recognition, and a touch panel 17 for inputting instructions from the user H.

The CPU 10 is a calculation device that controls each part connected by the bus line 13. The flash ROM 11 is a rewritable non-volatile memory that stores a voice recognition program 11a, all dictionary data 11b in which all dictionary data Ad is stored, and conversion dictionary data 11c in which conversion dictionary data Sd is stored. When the voice recognition program 11a is executed by the CPU 10, the voice processing of FIG. 4 is executed.

The conversion dictionary data 11c includes one-character dictionary data 11c1 in which the one-character dictionary data Sd1 described above is stored, two-character dictionary data 11c2 in which the two-character dictionary data Sd2 is stored, and similarly, three- to ten-character dictionary data 11c3 to 11c10 in which the three- to ten-character dictionary data Sd3 to Sd10 are stored, respectively. Hereinafter, the one- to ten-character dictionary data 11c1 to 11c10 are collectively referred to as "M-character dictionary data 11cM."

RAM 12 is a memory for rewritably storing various work data, flags, etc. when CPU 10 is executing voice recognition program 11a, and is provided with output character string memory 12a in which a word W obtained from input character string TiN is stored.

Next, the processing executed by the CPU 10 of the mobile terminal 1 will be described with reference to Figures 4 and 5. Figure 4(a) is a flowchart of the voice processing. The voice processing is a process that is executed when an instruction to execute the voice recognition program 11a is input from the user H via the touch panel 17 or the like.

The speech processing first checks whether the all dictionary data Ad of the all dictionary data 11b has been updated (S1). Specifically, the combinations of the pronunciation character strings Tp and words/phrases W of the all dictionary data Ad of the all dictionary data 11b are configured to be updated by adding or deleting the pronunciation character strings Tp or the words/phrases W, and it is checked whether an instruction to update the all dictionary data Ad of the all dictionary data 11b has been given by the user H via the touch panel 17.

In the process of S1, if the all dictionary data Ad of the all dictionary data 11b is updated (S1: Yes), since there is a possibility that the update of the all dictionary data Ad may cause a difference between the combination of the pronunciation character string Tp and the word W included in the all dictionary data Ad and the combination of the pronunciation character string Tp and the word W stored in the conversion dictionary data 11c, a conversion dictionary data creation process (S2) is executed. The conversion dictionary data creation process will be described with reference to FIG. 4(b).

Figure 4 (b) is a flowchart of the conversion dictionary data creation process. The conversion dictionary data creation process first obtains the updated all dictionary data Ad from the all dictionary data 11b (S20). After the process of S20, the combinations of the pronunciation character string Tp and the words and phrases W in the obtained all dictionary data Ad are rearranged in order of the number of characters in the pronunciation character string Tp (S21).

After the process of S21, the combination of the pronunciation string Tp and the word W is obtained for each number of characters of the pronunciation string Tp from the entire dictionary data Ad rearranged in order of the number of characters of the pronunciation string Tp to create dictionary data, and the dictionary data is stored in the M-character dictionary data 11cM of the corresponding number of characters in the conversion dictionary data 11c (S22). As a result, the M-character dictionary data 11cM of the conversion dictionary data 11c stores the same combination of the pronunciation string Tp and the word W as the entire dictionary data Ad of the updated entire dictionary data 11b.

After processing S22, the conversion dictionary data creation process ends.

Return to FIG. 4(a). If the entire dictionary data Ad of the entire dictionary data 11b has not been updated in the process of S1 (S1: No), the process of S2 is skipped. After the processes of S1 and S2, the output character string memory 12a is cleared (S3). After the process of S3, the voice V input from the microphone 15 is converted into a character string to obtain the above-mentioned input character string Ti (S4). Note that a well-known method is used for converting the voice V into a character string, and therefore a detailed description thereof is omitted.

After the process of S4, the acquired input character string Ti is subjected to morphological analysis to acquire an input character string TiN for each word (S5). Note that a known method is used for the morphological analysis, and therefore a detailed description thereof is omitted. After the process of S5, a dictionary application process (S6) is executed. The dictionary application process will be described with reference to FIG. 5.

Figure 5 is a flowchart of the dictionary application process. The dictionary application process first obtains the number of characters in each of the input strings TiN obtained in the process of S5 in Figure 4(a) (S30). After the process of S30, a counter variable N indicating the order of the input string TiN obtained from the input string Ti is set to 1 (S31). For example, when the counter variable N is "1", it is regarded as the above-mentioned "input string Ti1", and when the counter variable N is "2", it is regarded as the above-mentioned "input string Ti2". Hereinafter, the input string TiN at the Nth counter variable of the input string Ti is referred to as the "Nth input string TiN".

After the process of S31, M-character dictionary data 11cM corresponding to the number of characters of the N-th input character string TiN is obtained from the conversion dictionary data 11c (S32). After the process of S32, the N-th input character string TiN is searched for in the M-character dictionary data 11cM obtained in the process of S32, thereby obtaining a word W corresponding to the N-th input character string TiN (S33).

Specifically, the M-character dictionary data 11cM acquired in the process of S32 is searched for a pronunciation character string Tp that matches the Nth input character string TiN, and if a matching pronunciation character string Tp is found, the corresponding word or phrase W is acquired. On the other hand, if a pronunciation character string Tp that matches the Nth input character string TiN is not found in the M-character dictionary data 11cM, the Nth input character string TiN is acquired as it is as the word or phrase W.

After the process of S33, the obtained word W is added to the output character string memory 12a (S34). After the process of S34, 1 is added to the counter variable N (S35), and it is confirmed whether the counter variable N is greater than the number of input character strings TiN decomposed from the input character string Ti (S36). In the process of S36, if the counter variable N is equal to or less than the number of input character strings TiN (S36: No), the processes from S32 onwards are repeated. On the other hand, if the counter variable N is greater than the number of input character strings TiN (S36: Yes), the dictionary application process is terminated.

Returning to FIG. 4(a), after the dictionary application process of S6, the character string stored in the output character string memory 12a is displayed on the LCD 16 (S7). As a result, the voice V input by the user H is converted into an input character string Ti, and among the input character strings Ti, a character string that corresponds to a pronunciation character string Tp stored in the conversion dictionary data 11c is replaced with the corresponding word or phrase W and displayed on the LCD 16.

After the process of S7, it is confirmed whether an instruction to end the voice processing has been obtained from the user H via the touch panel 17 (S8). If an instruction to end the voice processing has not been obtained in the process of S8 (S8: No), the process from S1 onwards is repeated, and if an instruction to end the voice processing has been obtained (S8: Yes), the voice processing is terminated.

Next, a mobile terminal 100 according to a second embodiment will be described with reference to Figures 6 to 8. In the first embodiment described above, the conversion dictionary data Sd is created by dividing it into groups based on the number of characters in the pronunciation string Tp, and dictionary data (i.e., M-character dictionary data SdM) that matches the number of characters in the input string TiN is obtained from the conversion dictionary data Sd, and the word W is obtained using that dictionary data.

In contrast, in the second embodiment, conversion dictionary data Sd is created for each first character of the pronunciation string Tp, and dictionary data that matches the first character of the input string TiN is obtained from the conversion dictionary data Sd, and the word W is obtained using that dictionary data. The same parts as in the first embodiment described above are given the same reference numerals, and their description is omitted.

Figure 6(a) is a diagram that shows a schematic representation of the creation of conversion dictionary data Sd from all dictionary data Ad in the second embodiment, and Figure 6(b) is a diagram that shows a schematic representation of the replacement of an input character string Ti using conversion dictionary data Sd in the second embodiment. In the second embodiment, conversion dictionary data Sd is created for each first character of a pronunciation character string Tp from a combination of a pronunciation character string Tp and a word or phrase W in the all dictionary data Ad.

Specifically, the conversion dictionary data Sd of the second embodiment is constructed by dividing the combinations of the pronunciation character string Tp and the words and phrases W stored in the entire dictionary data Ad into the first characters of the pronunciation character string Tp. In the second embodiment, "hiragana" is used as the first character. That is, the conversion dictionary data Sd includes "a" dictionary data Sd20, which stores the combination of the pronunciation character string Tp whose first character is "a" and the words and phrases W corresponding to the pronunciation character string Tp, and "i" dictionary data Sd21, which stores the combination of the pronunciation character string Tp whose first character is "i" and the words and phrases W corresponding to the pronunciation character string Tp, and also includes "u" to "n" dictionary data Sd22 to 65, which store the combination of the pronunciation character string Tp whose first character is "u" to "n" and the words and phrases W corresponding to the pronunciation character string Tp. Hereinafter, the dictionary data Sd20 to 65 for "a" to "n" are collectively referred to as "P dictionary data SdP".

In the second embodiment, for combinations of a pronunciation string Tp and a word W in which a voiced consonant such as "ga" or a semi-voiced consonant such as "pi" is the first character, a character without these voiced consonants or semi-voiced consonants (for example, "ka" for "ga" and "hi" for "pi") is stored in the dictionary data for P SdP for the first character of the pronunciation string Tp. Note that this is not limited to this, and dictionary data for P SdP for combinations of a pronunciation string Tp and a word W in which a voiced consonant or semi-voiced consonant is the first character may be separately provided in the conversion dictionary data Sd.

The input character string Ti is replaced using the conversion dictionary data Sd configured in this way. In the example of FIG. 6(b), the input character string Ti is "ASAHAKUUKIGASUTERU", so by morphological analysis of this input character string Ti, the input character string Ti is decomposed into five words: an input character string Ti1 with the first character "A" based on "ASA", an input character string Ti2 with the first character "HA" based on "HA", an input character string Ti3 with the first character "KU" based on "KUUKI", an input character string Ti4 with the first character "GA" based on "GA", and an input character string Ti5 with the first character "SU" based on "SUTERU".

Then, for each decomposed input character string TiN, the conversion dictionary data Sd is referenced to obtain the word W. Specifically, the first character of each input character string TiN is obtained, and dictionary data SdP for P is obtained, which has a pronunciation character string Tp whose first character matches the obtained first character.

In the example of FIG. 2B, the first character of input string Ti1 is "a", so dictionary data Sd20 for "a" is obtained from the conversion dictionary data Sd, and a word or phrase W of a pronunciation string Tp that matches input string Ti1 is obtained from the dictionary data. Similarly, the first character of input string Ti2 is "ha", so dictionary data Sd45 for "ha" is obtained, and a word or phrase W is obtained from the dictionary data. The first character of input string Ti3 is "ku", so dictionary data Sd27 for "ku" is obtained, and a word or phrase W is obtained from the dictionary data. The first character of input string Ti4 is "ga", so dictionary data Sd25 for "ka" is obtained, and a word or phrase W is obtained from the dictionary data. The first character of input string Ti5 is "su", so dictionary data Sd32 for "su" is obtained, and a word or phrase W is obtained from the dictionary data.

In this way, dictionary data for P SdP is provided for each first character of the pronunciation string Tp, and each word or phrase W is searched for in the dictionary data for P SdP that matches the first character of the input string TiN. By searching for a word or phrase W in the dictionary data for P SdP that matches the first character of the input string TiN, it is possible to omit a comparison with the pronunciation string Tp, which has a number of characters that does not match the first character of the input string TiN in the first place, and therefore it is possible to quickly search for a word or phrase W.

In addition, since the first character of the pronunciation string Tp or the input string TiN can be easily obtained without complex analysis of the string, it is possible to quickly create the conversion dictionary data Sd and quickly obtain the dictionary data SdP for P that matches the first character of the input string TiN. This allows the process from when the user H inputs the voice V to when the word W corresponding to that voice V is displayed to be completed quickly.

Next, the electrical configuration of the mobile terminal 100 of the second embodiment will be described with reference to FIG. 7. FIG. 7 is a block diagram showing the electrical configuration of the mobile terminal 100 of the second embodiment. In the conversion dictionary data 11c of the mobile terminal 100 of the second embodiment, dictionary data 11c20-11c65 for "a" to "n" in which dictionary data Sd20-65 for "a" to "n" are stored, respectively, instead of M-character dictionary data 11cM. Hereinafter, dictionary data 11c20-11c65 for "a" to "n" will be referred to as "P dictionary data 11cP".

Next, referring to FIG. 8, the process executed by the CPU 10 of the mobile terminal 100 of the second embodiment will be described. FIG. 8(a) is a flowchart of the conversion dictionary data creation process of the second embodiment. After the process of S20, the conversion dictionary data creation process of the second embodiment rearranges the combinations of the pronunciation character string Tp and the words and phrases W in all the acquired dictionary data Ad in the order of the first characters of the pronunciation character string Tp (S100).

After the process of S100, the system obtains a combination of the pronunciation string Tp and the word W for each first character of the pronunciation string Tp from all dictionary data Ad rearranged in the order of the first characters of the pronunciation string Tp to create dictionary data, and stores the dictionary data in dictionary data 11cP for the corresponding first character P in the conversion dictionary data 11c (S101). After the process of S101, the conversion dictionary data creation process is terminated.

Next, the dictionary application process of the second embodiment will be described. FIG. 8(b) is a flowchart of the dictionary application process of the second embodiment. In the dictionary application process of the second embodiment, first, the first character of each input character string TiN obtained in the process of S5 described above in FIG. 4(a) is obtained (S110). After the process of S110, the process of S31 described above is performed, and after the process of S31, dictionary data 11cP for P corresponding to the first character of the Nth input character string TiN is obtained from the conversion dictionary data 11c (S111).

After the process of S111, the Nth input character string TiN is searched in the dictionary data for P 11cP obtained in the process of S111 to obtain a word W corresponding to the Nth input character string TiN (S112). The method of obtaining a word W corresponding to the input character string TiN from the dictionary data for P 11cP is the same as the process of S33 in FIG. 5 described above, so a detailed explanation is omitted. After the process of S112, the processes from S34 onwards are executed.

The present invention has been described above based on the embodiments, but the present invention is in no way limited to the above-mentioned embodiments, and it can be easily imagined that various improvements and modifications are possible within the scope of the invention without departing from its spirit.

In the first embodiment, conversion dictionary data Sd is created for each number of characters in the pronunciation string Tp, and in the second embodiment, conversion dictionary data Sd is created for each first character of the pronunciation string Tp, but this is not limited to the above. For example, conversion dictionary data Sd may be created for each combination of the number of characters and the first character of the pronunciation string Tp. In this case, the combination of the number of characters and the first character of the input string TiN is obtained, and dictionary data for the combination of the number of characters and the first character that corresponds to that combination is obtained from the conversion dictionary data Sd, and the obtained dictionary data is used to search for the word W.

This allows for omitting a comparison with pronunciation strings Tp that do not match the number of characters and first character of the input string TiN, making it possible to search for words and phrases W that correspond to the input string TiN even more quickly. Furthermore, by creating conversion dictionary data Sd for each combination of the number of characters and first character of the pronunciation string Tp, it is possible to reduce the number of pronunciation strings Tp and words and phrases W stored in one dictionary data, compared to the conversion dictionary data Sd for each number of characters in the first embodiment and the conversion dictionary data Sd for each first character in the second embodiment. This also allows for a quick search for words and phrases W that correspond to the input string TiN.

In addition, the conversion dictionary data Sd is not limited to being created according to the number of characters or the first character, but may be created according to other features of the pronunciation string Tp or the input string TiN. For example, the conversion dictionary data Sd may be created according to the sum of the character code values of the characters that make up the pronunciation string Tp or the input string TiN. In this case, the sum of the character code values of the characters that make up the input string TiN is obtained, dictionary data corresponding to that sum of character code values is obtained from the conversion dictionary data Sd, and the obtained dictionary data is used to search for the word W.

In the above embodiment, the total dictionary data Ad is provided, and the conversion dictionary data Sd is created from the total dictionary data Ad, but this is not limited to the above. For example, the total dictionary data Ad may be omitted, and the conversion dictionary data Sd may be configured only. In this case, instead of the processes S1 and S2 in the voice processing of FIG. 4, the conversion dictionary data Sd may be updated in response to an instruction from the user H.

In the above embodiment, the pronunciation string Tp and the input string TiN are composed of hiragana, but this is not limited to this and other phonetic characters such as katakana, the alphabet, or Chinese pinyin may be used. In addition, the conversion dictionary data Sd in the second embodiment is also created based on the first character in hiragana, but this is not limited to this and the conversion dictionary data Sd may be created based on the first character in other phonetic characters such as katakana, the alphabet, or Chinese pinyin.

In the first embodiment, the conversion dictionary data Sd includes 1-10 character dictionary data Sd3-10, which are up to 10 characters long, but this is not limited to this and dictionary data of 10 characters or more may be provided.

In the above embodiment, the input character string Ti is decomposed into word-specific input character strings TiN, and each of the input character strings TiN is searched in the conversion dictionary data Sd, but this is not limited thereto, and the conversion dictionary data Sd may be searched for an input character string Ti that includes multiple words.

In the above embodiment, the decomposition of the input character string Ti into the input character string TiN for each word is performed by morphological analysis, but this is not limited to this, and the decomposition of the input character string Ti into the input character string TiN for each word may be performed by other methods such as AI.

In the above embodiment, the character string in the output character string memory 12a is displayed on the LCD 16 in the process of S34 in Figs. 5 and 8(b), but this is not limited to this. For example, the character string in the output character string memory 12a may be transmitted to an information processing device such as another mobile terminal 1 via a communication device (not shown), or a printer (not shown) may be connected and the character string in the output character string memory 12a may be printed on paper.

In the above embodiment, a mobile terminal 1 incorporating the voice recognition program 11a is exemplified, but the present invention is not limited to this, and the voice recognition program 11a may be executed by another information processing device (computer) such as a personal computer or a tablet terminal. In addition, the voice recognition program 11a may be stored in a ROM or an IC chip, and the present invention may be applied to a dedicated device that executes only the voice recognition program 11a.

1,100 Mobile terminal (computer)
11 Flash ROM (storage unit)
11a Voice recognition program V Voice Tp Pronounced character string W Word Sd Conversion dictionary data 11c Conversion dictionary data (conversion dictionary storage means)
Ti, TiN Input character string S4 Speech conversion step, speech conversion means S30, S110 Feature acquisition step, feature acquisition means S32, S111 Dictionary acquisition step, feature acquisition means S33, S112 Search step, search means S7 Output step, output means

Claims (6)

A speech recognition program for causing a computer having a storage unit to execute a speech recognition process,
The storage unit is caused to function as a conversion dictionary storage means for storing a plurality of conversion dictionary data, which are dictionary data consisting of a combination of a pronunciation character string and a word corresponding to the pronunciation character string, and which are divided by feature amount of the pronunciation character string;
a speech conversion step for converting the input speech into an input string representing the pronunciation of the input speech;
a feature acquisition step of acquiring features of the input character string converted in the speech conversion step;
a dictionary acquisition step of acquiring conversion dictionary data corresponding to the feature amount acquired in the feature amount acquisition step from the conversion dictionary storage means;
a search step of searching for a phrase corresponding to the input character string converted in the speech conversion step from the conversion dictionary data acquired in the dictionary acquisition step;
an output step of outputting the searched word or phrase when the word or phrase is found in the search step , and outputting the input character string converted in the speech conversion step when the word or phrase is not found in the search step,
the conversion dictionary data is created from all dictionary data in which combinations of all pronunciation character strings included in the plurality of conversion dictionary data and words corresponding to the pronunciation character strings are stored;
A speech recognition program characterized in that, when the entire dictionary data is updated, combinations of pronunciation strings and words corresponding to those pronunciation strings contained in the updated entire dictionary data are rearranged based on features of the pronunciation strings, and conversion dictionary data for each feature of the pronunciation strings is created from the rearranged combinations of pronunciation strings and words corresponding to those pronunciation strings . The speech conversion step converts the input speech into a character string representing the pronunciation, and breaks the character string into words to obtain an input character string;
The feature amount acquiring step acquires feature amounts of each of the input character strings converted and decomposed into words in the speech conversion step,
The dictionary acquisition step acquires, from the conversion dictionary storage means, conversion dictionary data corresponding to each feature of the input character string decomposed into words acquired in the feature acquisition step,
2. The speech recognition program according to claim 1, wherein the search step searches for a phrase corresponding to each of the input character strings decomposed into words converted in the speech conversion step from each conversion dictionary data acquired in the dictionary acquisition step. The speech recognition program according to claim 1 or 2, characterized in that the feature is the number of characters in the pronunciation string or the input string. The speech recognition program according to claim 1 or 2, characterized in that the feature is the first character of the pronunciation string or the input string. The speech recognition program according to claim 1 or 2, characterized in that the feature is a combination of the number of characters and the first character of the pronunciation string or the input string. a conversion dictionary storage means for storing a plurality of conversion dictionary data, the conversion dictionary data being a combination of a pronunciation character string and a phrase corresponding to the pronunciation character string, the conversion dictionary data being classified according to features of the pronunciation character string;
A voice input means for inputting voice;
a speech conversion means for converting the speech inputted by the speech input means into an input character string representing the pronunciation of the speech;
A feature acquisition means for acquiring a feature of an input character string converted by the speech conversion means;
a dictionary acquisition means for acquiring conversion dictionary data corresponding to the feature acquired by the feature acquisition means from among the conversion dictionary data stored in the conversion dictionary storage means;
a search means for searching conversion dictionary data acquired by the dictionary acquisition means for a word or phrase corresponding to the input character string converted by the speech conversion means;
an output means for outputting the searched word or phrase when the search means is able to search for the word or phrase , and for outputting the input character string converted by the speech conversion means when the search means is unable to search for the word or phrase,
the conversion dictionary data is created from all dictionary data in which combinations of all pronunciation character strings included in the plurality of conversion dictionary data and words corresponding to the pronunciation character strings are stored;
A speech recognition device characterized in that, when the entire dictionary data is updated, combinations of pronunciation strings and words corresponding to the pronunciation strings contained in the updated entire dictionary data are rearranged based on features of the pronunciation strings, and conversion dictionary data for each feature of the pronunciation strings is created from the rearranged combinations of pronunciation strings and words corresponding to the pronunciation strings .

Images