JP2006209077A - Voice interactive device and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To make word chains with a speaker by speech-recognizing a vocally inputted word and outputting a word starting with the same character as the ending character of the word recognized. <P>SOLUTION: A controller 106 voice-recognizes the beginning and the ending of a word inputted via a microphone 101, determines whether the inputted word have been spoken, based on a making-word chains rule, based on the speech-recognized beginning and ending of the word, and extracts an answering word, which starts with the same character with the ending of the speech-recognized word, when it is determined that the inputted word has been spoken, according to the making-word chains rule and vocally output the word via a loudspeaker 105. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a speech dialogue apparatus and method for recognizing a word spoken by a speaker and recognizing it as a speaker.

  The following shiritori game machine is known from Patent Document 1. In this shiritori game machine, after the speech recognition is performed by comparing the word uttered by the speaker with the word stored in the word storage means, the beginning and ending characters of the speech recognized word are extracted. Then, a word that starts with the speech-recognized ending and does not end with “n” is read from the word storage means and output as speech.

JP-A-6-269534

  However, in the conventional shiritori game machine, since the beginning and ending of the word are extracted after speech recognition of the word spoken by the speaker, it is necessary to recognize the entire word once, and the number of words stored in the word storage means is As the number of times increases, the time required for speech recognition increases, and the problem of an increased recognition rate has arisen.

  The present invention relates to a spoken dialogue apparatus and method for performing conversation with a speaker, extracting a word head and a word ending from a word input via a voice input means, performing voice recognition, and applying the word recognition to the word head and word ending. If it is determined that the input word is uttered according to Shiritori rules, and the input word is uttered according to Shiritori rules, The response word starting from the same character as the end of the speech-recognized word is extracted and output as speech.

  According to the present invention, the beginning and ending of the input word are recognized by voice. As a result, taking into account that the beginning and end of a word are used for rule determination in shiritori, the processing speed is improved and the misrecognition rate is reduced by recognizing only the part that becomes the point. be able to.

-First embodiment-
FIG. 1 is a block diagram showing a configuration of an embodiment of a voice interaction apparatus according to the first embodiment. The voice interactive apparatus 100 includes a microphone 101 for inputting a word spoken by a speaker, a voice input operation switch 102 for instructing start of voice input, that is, start of shiritori, and a word input via the microphone 101. A recognition syllable dictionary 103 that stores all 50 syllables of “a” to “n” as standby syllables for recognizing the beginning and ending, and a response word database 104 that stores response words for shiritori. A speaker 105 for outputting a response word to a speaker as a voice, a control device 106 for executing various processes such as recognition of a prefix of an input word and determination of a response word, and a past shiritori History, that is, a history for storing speech data history of words spoken by the speaker and response words of the spoken dialogue apparatus 100 responding to the speaker And a memory 107.

  When the voice data of a word is input via the microphone 101, the control device 106 first extracts the beginning and ending of the input word. That is, as shown in FIG. 2, the part 2a corresponding to the beginning of the word and the part 2b corresponding to the ending are extracted from the input voice data. Then, the extracted beginning and ending speech data and the standby syllables stored in the recognition syllable dictionary 103 are collated to calculate the recognition likelihood of the beginning and ending in each standby syllable, that is, the probability. . Then, standby syllables with the maximum recognition likelihood are extracted for each of the beginning and ending, and each is used as a recognition result candidate for the beginning and ending.

  In the first embodiment, the recognition likelihood is represented by a numerical value of 0 to 1, for example, 0 when the extracted head or ending and the standby syllable do not match at all. 1 is calculated. That is, it is assumed that the recognition likelihood increases as the coincidence between the beginning and ending and the standby syllable increases. For example, when an “apple” is uttered by a speaker and a voice is input, as shown in FIG. 3, the recognition result candidate at the beginning is “ri” and the recognition result candidate at the ending is “go”. It is assumed that (A), for example, 0.7, and (B), for example, 0.8 are calculated as the recognition likelihood at that time. In this case, the response content to the speaker is changed based on the calculated recognition likelihood of the beginning and ending.

  That is, it is determined whether the recognition likelihood of the beginning and the recognition likelihood of the ending are each greater than a predetermined value set in advance. Here, the predetermined value is set such that if the likelihood of recognition is greater than the predetermined value, the degree of coincidence between the beginning and ending and the standby syllable is sufficiently high and can be adopted as a speech recognition result. . Then, the comparison result between the recognition likelihood of the beginning and the end of the word and this predetermined value is classified into four patterns 1 to 4 shown in FIG. 4, and as shown in the following (1) to (4) The response content to the speaker is changed for each pattern.

(1) Pattern 1: When the recognition likelihood of the beginning and the ending is both larger than a predetermined value In this case, the ending of the word spoken by the speaker is specified because the recognition likelihood of the ending is larger than the predetermined value. Can do. Therefore, it is possible to output a response word starting from the same character as this ending. However, since it is necessary to determine whether the word uttered by the speaker is in accordance with the general rules of shiritori, the following processing is performed. In addition, as a general rule of shiritori, “the ending of the spoken word is not“ n ””, “whether the beginning of the spoken word begins with the same character as the ending of the previous response word”, and “ Whether or not a word that appeared in the past has been uttered repeatedly is determined.

  For this purpose, first, it is determined whether or not the ending that can be identified as a result of speech recognition is “n”. In addition, it is determined whether or not the recognition result candidate for the beginning of the word matches the ending of the response word responded by the spoken dialogue apparatus 100 immediately before being stored in the history memory 107. If it is determined that the word spoken by the speaker is not in accordance with the rules of shiritori based on the result of the rule determination (first rule determination) so far, “Please speak again” to the speaker A response message that prompts the user to repeat a new word is output via the speaker 105.

  On the other hand, regarding the result of the first rule determination so far, when it is determined that the word uttered by the speaker is in accordance with the shiritori rule, the second rule determination is further executed and “past It is determined whether or not the word that appears in is repeatedly uttered. In the first embodiment, as described above, only the beginning and ending of the uttered word are extracted and speech recognition is performed, so the second rule determination is performed using only the beginning and ending of the word. For this reason, among all the words that have appeared in the past included in the shiritori history stored in the history memory 107, that is, the pair of the beginning and ending in the response word that has already been spoken by the speaker and the response word that has been output by the voice interactive device 100 In addition, it is determined whether or not there are a predetermined number or more of coinciding word pairs that are extracted from words spoken by the speaker.

  It has been determined that there are more than a predetermined number of pairs of word prefixes and endings of all the words that have appeared in the past stored in the history memory 107 that match the word prefix and ending pairs of words spoken by the speaker this time. In this case, the speaker determines that the word that has already appeared has been uttered repeatedly. Then, a response message prompting the re-speech of a word such as “The word has already been spoken. Please utter another word” is output to the speaker via the speaker 105.

  In contrast, a predetermined number of pairs of beginning and ending words of all the words that have appeared in the past stored in the history memory 107 coincide with the beginning and ending pairs of the words spoken by the speaker this time. If it is determined that there is no more, a word spoken by the speaker is accepted, an arbitrary response word starting from the recognized ending and not existing in the history memory 107 is extracted from the response word database 104, and the speaker The data is output via 105. At this time, if there is no corresponding response word in the response word database 104, the speaker wins.

  It should be noted that in such a method for determining an already-occurring word, a word having the same combination of beginning and ending has appeared more than a predetermined number of times in the past, even though the speaker actually uttered a word that has not yet appeared. If the number of occurrences of the word with the same combination of beginning and end is less than the predetermined number of times, it is determined that the word does not comply with the rule. It may be determined that there is. However, in the first embodiment, since only the beginning and ending of the spoken word are extracted and speech recognition is performed, the rule determination is performed uniformly considering only the combination of the beginning and ending of the word. And

(2) Pattern 2: When the recognition likelihood of the beginning is less than a predetermined value and the recognition likelihood of the ending is larger than the predetermined value In this case, since the recognition likelihood of the ending is larger than the predetermined value, the speaker speaks The ending of the word can be specified. For this reason, the same processing as that of the pattern 1 described above is performed.

(3) Pattern 3: When the recognition likelihood of the beginning is larger than a predetermined value and the recognition likelihood of the ending is less than or equal to a predetermined value In this case, since the recognition likelihood of the ending is less than or equal to a predetermined value, the speaker speaks It is impossible to specify the ending of the selected word and to determine the response word to be output to the speaker. Therefore, a response message that prompts the speaker to re-speak a word such as “Please speak again” is output via the speaker 105.

(4) Pattern 4: When the recognition likelihood of the beginning and the ending is both equal to or less than the predetermined value In this case, since both the beginning and the ending cannot be normally recognized, the speaker can immediately The output response word may not be understood correctly and may not be spoken clearly. Therefore, a response message for prompting the re-speaking by presenting the immediately previous response word to the speaker again is output via the speaker 105. For example, if the immediately preceding response word stored in the history memory 107 is “papaya”, a response message such as “Please think about papaya“ ya ”” is output.

  FIG. 5 is a flowchart showing processing of the voice interactive apparatus 100 in the first embodiment. The processing shown in FIG. 5 is executed by the control device 106 when the voice interactive device 100 is turned on. In step S10, it is determined whether or not the voice input operation switch 102 has been pressed by the speaker and the start of shiritori has been instructed. If it is determined that the voice input operation switch 102 has been pressed, the process proceeds to step S20. In step S <b> 20, an arbitrary response word stored in the response word database 104 is extracted, and the sound is output from the speaker 104. Then, it progresses to step S30.

  In step S30, the output voice data of the response word is stored in the history memory 106, and the process proceeds to step S40. In step S <b> 40, it is determined whether or not a word is uttered by the speaker and input via the microphone 101. If it is determined that a word has been uttered by the speaker, the process proceeds to step S50. In step S50, the part 2a corresponding to the beginning of the word and the part 2b corresponding to the ending are extracted from the voice data of the input word, and the process proceeds to step S60. In step S60, the extracted beginning and ending speech data and the standby syllable stored in the recognition syllable dictionary 103 are collated to calculate the recognition likelihood of the beginning and ending in each standby syllable. Thereafter, the process proceeds to step S70.

  In step S70, it is determined whether or not the calculated ending recognition likelihood is greater than the predetermined value. If it is determined that the ending recognition likelihood is greater than the predetermined value, the process proceeds to step S80. In step S80, since it corresponds to the above-described pattern 1 or pattern 2, as the first rule determination, “whether the utterance of the spoken word is not“ n ”” and “the head of the spoken word is the previous response” It is determined whether or not it begins with the same character as the end of the word. Then, the process proceeds to step S90, and the process branches based on the result of the first rule determination.

  If it is determined that the word spoken by the speaker is not in accordance with the first rule, the process proceeds to step S130, and a response message that prompts the speaker to re-utter the word is output via the speaker 105. On the other hand, if it is determined that the word spoken by the speaker conforms to the first rule, the process proceeds to step S100. In step S100, in order to determine “whether the word that has appeared in the past has been repeatedly uttered” as the second rule determination, the pair of the beginning and the ending of all the words that have been stored in the history memory 107 in the past. Then, it is determined whether or not there are a predetermined number or more that match the pair of the beginning and ending of the word spoken by the speaker. And it progresses to step S110 and branches a process based on the determination result.

  When it is determined that the speaker has repeatedly uttered a word that has already appeared, the process proceeds to step S130 described above, and a response message that prompts the speaker to re-utter the word is output via the speaker 105. On the other hand, if it is determined that the speaker has not repeatedly spoken words that have already appeared, the process proceeds to step S150. In step S150, a word uttered by the speaker is accepted, an arbitrary response word starting from the recognized ending and not existing in the history memory 107 is extracted from the response word database 104, and the process proceeds to step S160.

  In step S160, it is determined whether or not the response word has been extracted from the response word database 104. If it is determined that the response word can be extracted, the chatter can be continued, and the process proceeds to step S170, where the speech data of the word input by the speaker is stored in the history memory 107. Thereafter, the process returns to step S20, and the extracted response word is output via the speaker 105. On the other hand, if it is determined that the response word cannot be extracted from the response word database 104, the process proceeds to step S180, where it is determined that the user has won, and the process ends.

  Next, processing when it is determined in step S70 that the ending recognition likelihood is equal to or less than a predetermined value will be described. In this case, the process proceeds to step S120. In step S120, it is determined whether the recognition likelihood of the beginning of the word is greater than a predetermined value. If it is determined that the recognition probability of the beginning of the word is larger than the predetermined value, the pattern 3 corresponds to the above-described pattern 3, and thus the process proceeds to step S130, and a response message that prompts the speaker to re-utter the word is sent to the speaker 105. To output. In contrast, if it is determined that the recognition likelihood of the beginning of the word is equal to or less than a predetermined value, the process proceeds to step S140. In this case, since it corresponds to the pattern 4 described above, a response message for prompting a recurrence speech by presenting the immediately previous response word to the speaker again is output via the speaker 105.

According to the first embodiment described above, the following operational effects can be obtained.
(1) Only the beginning and ending of a word input by a speaker are extracted, and the extracted word is collated with a standby syllable stored in the recognition syllable dictionary 103, so that the beginning and ending of each standby syllable are detected. The recognition likelihood is calculated and voice recognition is performed. As a result, the target of speech recognition can be limited to only the beginning and ending of the ritual point, and the load of recognition processing can be reduced.

(2) In addition, since there is no need to recognize characters contained between the beginning and end of a word, it is not necessary to prepare a large number of words as standby words at the time of speech recognition. It becomes possible to respond.

(3) When the recognition likelihood of the ending is less than or equal to a predetermined value, the utterer is prompted to recite. Thereby, erroneous recognition can be prevented.

(4) When both the initial and final recognition likelihoods are less than or equal to a predetermined value, the immediately preceding response word is presented to the speaker to encourage recurrent speech. As a result, when there is a possibility that the speaker does not understand the immediately preceding response word, by re-presenting the immediately preceding response word, information on the beginning of the speech to be spoken by the speaker is presented so that the conversation can be performed smoothly. You can continue.

(5) As the first rule determination, it is determined whether or not the ending that can be specified as a result of the speech recognition is “n”, and the speech dialogue apparatus 100 immediately before the initial recognition result candidate is stored in the history memory 107. Judgment is made as to whether or not it matches the end of the response word. Thereby, it is possible to easily determine whether or not a word spoken by a speaker is in accordance with a general rule of shiritori.

(6) There are more than a predetermined number of pairs of beginning and ending parts of all words that have appeared in the past stored in the history memory 107 that match the beginning and ending pairs of the words spoken by the speaker this time. If so, the speaker determines that the word that has already appeared has been uttered repeatedly. As a result, it is not necessary to perform matching processing on the entire word spoken by the speaker and all of the words that have appeared in the past stored in the history memory 107, and only the combination of the beginning and the end of the word needs to be matched. Therefore, the load of the determination process can be reduced.

-Second embodiment-
In the first embodiment described above, the response word for shiritori is stored in the response word database 104 and stored in the response word database 104 based on the words spoken by the speaker. The example in which the response word is extracted and output via the speaker 105 has been described. On the other hand, in the second embodiment, in the response word database 104, response words are classified and stored in categories, and a difficulty level (level) is added to each response word.

  The user can control a word output from the voice interactive apparatus 100 by designating a category as a range of words used in shiritori and further setting a level of difficulty of shiritori (adjusting the level). .

  In the second embodiment, a diagram showing a specific example of extracting the word head and ending based on the waveform of the speech voice data shown in FIG. 2, and the word head and ending and the standby shown in FIG. About each figure of the figure which shows the specific example of the collation result with a syllable, and the figure which shows the comparison result with the recognition likelihood of the beginning and ending shown in FIG. 4, and a predetermined value, it is the same as that of 1st Embodiment Therefore, the description is omitted.

  Further, in the following description, as a result of performing the above-described shiritori rule determination in the first embodiment for a word input by the user via the microphone 101, the word spoken by the speaker is shiritori. It is assumed that it is determined that it conforms to the rules of

  FIG. 6 is a block diagram illustrating a configuration of an embodiment of a voice interaction apparatus according to the second embodiment. In FIG. 6, the same components as those in the first embodiment shown in FIG. The voice interactive apparatus 100 further includes a monitor 108 that displays response words output to the speaker, various menus, and the like.

  The monitor 108 includes a touch panel 108a operated by a user. The user can select any touched item by touching (touching) any item on the menu displayed on the monitor 108 and instruct the voice interactive apparatus 100 to execute the process. . In this embodiment, an example in which the user operates the touch panel 108a to input a command to the voice interactive device 100 will be described. However, other input devices such as a remote controller and a hard switch are installed, and the user A command may be input via the input device. In addition, a voice command may be input via the microphone 101.

  FIG. 7 is a diagram schematically showing response words stored in the response word database 104 according to the second embodiment. Here, a case will be described in which it is determined that the ending of a word uttered by the user is “A”. Therefore, in the example shown in FIG. 7, only the response word starting with “A” is shown.

  As shown in FIG. 7, each response word has a label (headword) 7a indicating the reading of the word, a part of speech 7b of the response word, a category 7c to which the response word belongs, and a level 7d described later. It is associated.

  The category 7c is information on the range of response words used for shiritori, and shows the result of classifying each response word such as a station name, place name, plant, nature, or animal into a category according to its meaning. In order to limit the range of words used in shiritori by the user, when “station name” is designated as the category, only the response word whose label 7a is “station name” and whose category 7c is “station name” is used. Extract as response word candidates.

  Level 7d is information on the difficulty level of shiritori and is represented by a numerical value corresponding to the difficulty level of shiritori. There are three levels of difficulty of shiritori: “easy”, “ordinary”, and “difficult”, and the easier the difficulty is, the fewer words that the voice interactive device 100 can respond to, and the possibility that the user will win To increase. On the other hand, as the difficulty level is “difficult”, the number of words that can be answered by the voice interaction apparatus 100 is increased and decreased to reduce the possibility that the user will win. In this embodiment, “1” is set as the level 7c corresponding to the difficulty level “easy”, and “2” is set as the level 7c corresponding to the difficulty level “ordinary”, which corresponds to the difficulty level “difficult”. “3” is set as the level 7c.

  When the difficulty level is set to “easy” by the user, in order to reduce the number of words that the voice interaction apparatus 100 can respond to, only the response word with level “1c” of “1” is used as a word for shiritori. . When the difficulty level is set to “ordinary” by the user, in order to increase the number of words that can be answered by the spoken dialogue apparatus 100 as compared with the case where the difficulty level is “easy”, the response word having the level 7c of “1” Then, a response word having a level 7c of “2” is used as a word used for shiritori. When the difficulty level is set to “difficult” by the user, in order to increase the number of words that the voice interaction apparatus 100 can respond to, all the response words whose level 7c is “1” to “3” are displayed. It is a word used in shiritori.

  Hereinafter, a specific example will be described in which the category of the response word and the difficulty level of shiritori are set by the user in order to control the word output from the voice interaction apparatus 100 as described above. FIG. 8 is a diagram showing a specific example of a setting screen displayed on the monitor 108 for the user to specify a category of response words and to set a difficulty level of shiritori. The user operates the touch panel 108a to set in advance the response word category and the level of difficulty of shiritori as the response word extraction conditions on the setting screen shown in FIG.

  FIG. 8 shows a specific example in which “difficult” is set in the difficulty setting 8a as the difficulty level of the shiritori, and “station name” is specified in the category setting 8b as the category of the response word. In this way, when the response word category and the shiritori difficulty level are set on the setting screen shown in FIG. 8, the control device 106 sets the set response word category and the shiritori difficulty level. Is used as a response word extraction condition, only response words that match the extraction condition are extracted as matching words from the response words stored in the response word database 104 described above with reference to FIG.

  That is, in the example shown in FIG. 8, since “difficult” is designated by the difficulty setting 8a by the user and “station name” is designated by the category setting 8b, the control device 106 responds as described above with reference to FIG. From the response words stored in the service word database 104, all response words whose category 7c is “station name” and whose level 7c is “1” to “3” are extracted as matching words. As a result, from the response words shown in FIG. 7, only the response word whose label 7a is “Akiru” is extracted as a matching word.

  In the response word database 104, when there is no response word that matches the extraction condition set by the user on the setting screen shown in FIG. 8, that is, there is no response word extracted as a matching word. In that case, the speaker wins. And the control apparatus 106 outputs the guidance which declares losing via the speaker 105, for example, "There is no word which can be thought. I am losing."

  If one or more response words are extracted as matching words as a result of the processing described above, the extracted response word is not the same as the word already spoken by the speaker, and the voice interaction apparatus 100 Is not the same as the already responded word. For this reason, the processing in the second rule determination performed when determining whether or not the speaker has already uttered a word that has already appeared in the first embodiment is performed on each extracted matching word. Execute. That is, the matching word is already used by comparing the pair of the beginning and ending of each extracted matching word with the pair of the beginning and ending of all the words that have been stored in the history memory 107 in the past. Determine whether it is a word. Then, response words that are already used are excluded from the extracted matching words.

  As a result, if the number of remaining matching words is 0, the speaker wins, and the control device 106 outputs the guidance for declaring the loss as described above via the speaker 105. On the other hand, if there is one remaining matching word, the remaining matching word is output through the speaker 105 and responded. If there are two or more matching words remaining, any one word is randomly extracted from the remaining matching words, and is output via the speaker 105 to respond.

  FIG. 9 is a flowchart showing processing of the voice interaction apparatus 100 according to the second embodiment. The processing shown in FIG. 9 is executed by the control device 106 when the voice interaction device 100 is powered on. In FIG. 9, the same processing steps as those of the voice interactive device 100 in the first embodiment shown in FIG. 5 are given the same step numbers, and differences will be mainly described. In step S151, the response word extraction process shown in FIG. 10 is executed.

  FIG. 10 is a flowchart showing a response word extraction process in the second embodiment. In step S210, the extraction condition of the response word preset by the user, that is, the set category of response word and the difficulty level of the designated shiritori are read on the setting screen described above with reference to FIG. Thereafter, the process proceeds to step S220, and based on the set extraction condition, only the response word that matches the extraction condition is selected from the response words stored in the response word database 104 described above with reference to FIG. Extract as Thereafter, the process proceeds to step S230.

  In step S230, it is determined whether a matching word has been extracted. If it is determined that no matching word is extracted, it is determined that the response word cannot be extracted, and the process returns to the process shown in FIG. On the other hand, if the extracted word is extracted, the process proceeds to step S240. In step S240, the beginning and ending of the extracted matching word are extracted, and the process proceeds to step S250. In step S250, similar to the processing in the second rule determination in the first embodiment, the pair of the beginning and end of each extracted matching word and all the words that have appeared in the past stored in the history memory 107 are displayed. The beginning and ending pairs are compared to determine whether each matching word is already used.

  Thereafter, the process proceeds to step S260, where the already used response word is excluded from the extracted matching words, and the process proceeds to step S270. In step S270, the number of matching words remaining as a result of excluding the response words already used is determined. If it is determined that there are no remaining matching words, it is determined that the response word cannot be extracted, and the process returns to the process illustrated in FIG.

  On the other hand, if it is determined that there is one remaining matching word, the process proceeds to step S280, and the remaining matching word is determined as a response word. Thereafter, the process returns to the process shown in FIG. If it is determined that there are a plurality of remaining matching words, the process proceeds to step S290, where one arbitrary word is randomly extracted from the remaining matching words and determined as a response word. Thereafter, the process returns to the process shown in FIG.

According to the second embodiment described above, the following effects can be obtained in addition to the operational effects of the first embodiment.
(1) From the response words stored in the response word database 104, only the response words in the category that matches the category of the response word set by the user are extracted as matching words and used. Word candidates to respond to the person. As a result, it is possible to perform a shiritori for only words of a specific category (genre).

(2) Matching words extracted according to the degree of difficulty specified by the user from among the response words stored in the response word database 104 are used as word candidates to respond to the user. I made it. Thus, the user can arbitrarily set the difficulty level of the shiritori game, and can provide a shiritori game for a wide range of users from advanced users to beginners.

-Modification-
Note that the voice interaction apparatus according to the embodiment described above can be modified as follows.
(1) In the first and second embodiments described above, the recognition likelihood is represented by a numerical value of 0 to 1, and the beginning and ending extracted from the words spoken by the speaker and the standby syllables do not coincide at all. The example in which 0 is calculated in the case and 1 is calculated in the case of complete coincidence has been described. However, the present invention is not limited to this, and the recognition likelihood may be calculated by other calculation methods.

(2) In the first and second embodiments described above, the beginning extracted from the word spoken by the speaker in the pair of the beginning and ending of all the words that have been stored in the history memory 107 in the past An example has been described in which it is determined that a speaker has repeatedly uttered a word that has already appeared when it is determined that there is a predetermined number or more matching the pair of ending and ending. However, the present invention is not limited to this, and the waveform of the speech data of the words spoken by the speaker and the waveform of the speech data of all the words that have already been stored stored in the history memory 107 are compared and stored in the history memory 107. If there is a word whose waveform is similar to the word spoken by the speaker, it may be determined that the speaker has repeatedly spoken a word that has already appeared. Further, in the second embodiment, the same applies to the case where it is determined whether or not the extracted matching word is a used word.

(3) In the above-described second embodiment, an example has been described in which the user designates a word category in order to designate a word range used in shiritori. However, the present invention is not limited to this. For example, the range of words used in shiritori may be designated by specifying the part of speech 7b shown in FIG. 7, and a condition for designating the range of words used in other shiritori can be set. It may be. In this case, information serving as a condition for designating a range of words used in shiritori is added to the response word stored in the response word database 104.

  Note that the present invention is not limited to the configurations in the above-described embodiments as long as the characteristic functions of the present invention are not impaired.

  The correspondence between the constituent elements of the claims and the embodiment will be described. The microphone 101 corresponds to voice input means, the response word database 104 corresponds to storage means, and the history memory 107 corresponds to history storage means. The control device 106 corresponds to speech recognition means, rule determination means, output means, and appeared word determination means. The touch panel 108a corresponds to a difficulty level setting unit and a range setting unit. This correspondence is an example, and the correspondence is different depending on the configuration of the embodiment.

It is a block diagram which shows the structure of one Embodiment of a voice interactive apparatus. It is a figure which shows the specific example which extracts the beginning and ending of a word based on the waveform of the audio | voice data which the speaker uttered. It is a figure which shows the specific example of the collation result with the beginning and ending of a word, and a standby syllable. It is a figure which shows the comparison result of the recognition likelihood of a head part and a word end, and a predetermined value. It is a flowchart figure which shows the process of the voice interactive apparatus 100 in 1st Embodiment. It is a block diagram which shows the structure of one Embodiment of the voice interactive apparatus in 2nd Embodiment. It is the figure which showed typically the word for response stored in the word database for response 104 in 2nd Embodiment. It is a figure which shows the specific example of the setting screen for performing category designation | designated of the word for a response, and setting the difficulty level of shiritori. It is a flowchart figure which shows the process of the voice interactive apparatus 100 in 2nd Embodiment. It is a flowchart figure which shows the response word extraction process in 2nd Embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Voice interaction apparatus 101 Microphone 102 Voice input operation switch 103 Recognition syllable dictionary 104 Response word database 105 Speaker 106 Control apparatus 107 History memory 108 Monitor 108a Touch panel

Claims (10)

A spoken dialogue device that interacts with a speaker,
Speech recognition means for recognizing speech by extracting the beginning and ending from the word input via the speech input means;
Storage means for storing a response word for shiritori;
Rule determining means for determining whether or not the input word is uttered in accordance with a rule of shiritori based on the beginning and ending of the word recognized by the voice recognition means;
When it is determined that the word input by the rule determination unit is uttered according to a rule of shiritori, the response word starting from the same character as the ending of the word recognized by the voice recognition unit is A voice dialogue apparatus comprising: output means for extracting voice from the storage means and outputting the voice. The voice interactive apparatus according to claim 1,
The speech recognition means extracts the beginning and ending from the speech data of the word input via the speech input means, and compares the extracted beginning and ending with the standby syllable to calculate each recognition likelihood. A speech dialogue apparatus characterized by recognizing the beginning and ending of a word by means of voice. The voice interaction apparatus according to claim 2,
A spoken dialogue apparatus characterized by prompting a speaker to recite a word when the ending recognition likelihood calculated by the voice recognition means is a predetermined value or less. The voice interaction apparatus according to claim 2 or 3,
When the speech recognition likelihood of the beginning calculated by the speech recognition means is less than or equal to a predetermined value and the recognition likelihood of the ending is also less than or equal to a predetermined value, information on the beginning of the speaker to be uttered by the speaker is presented to the speaker A speech dialogue apparatus characterized by prompting a re-speech of a word. In the voice interactive apparatus according to any one of claims 1 to 4,
A history storage means for storing the history of the word input via the voice input means and the response word output via the output means as a slicing history;
The rule determination means refers to the bookmarking history stored in the history storage means and determines whether or not the word input via the voice input means is an already appearing word Including
When the appearing word determination means determines that the word input via the voice input means is an already appearing word, it prompts the speaker to recite the word. Spoken dialogue device. The voice interaction apparatus according to claim 5, wherein
The appearance word determination means includes a pair of the beginning and ending of the word speech-recognized by the speech recognition means in the pair of the beginning and ending of all the words included in the shiritori history stored in the history storage means. A spoken dialogue apparatus characterized in that if a predetermined number or more of the same ones exist, it is determined that a word input through the voice input means has already appeared. The voice interaction apparatus according to claim 5, wherein
The appearance word determination unit is similar to the waveform of the speech data of the word input via the speech input unit in the waveform of the speech data of all the words included in the shiritori history stored in the history storage unit. A spoken dialogue apparatus characterized in that, when a word exists, it is determined that the word input through the voice input means has already appeared. In the voice interactive apparatus according to any one of claims 1 to 7,
It further comprises a difficulty level setting means for setting the difficulty level of the shiritori.
Information about the difficulty level of each shiritori is added to each of the shiritori response words stored in the storage means,
The output means extracts a response word corresponding to the difficulty level of the shiritori set by the difficulty level setting means from the storage means based on the information on the difficulty level of the shiritori, and outputs it as a voice. Voice dialogue device. In the voice interaction device according to any one of claims 1 to 8,
It further comprises range setting means for setting a range of response words used for shiritori,
Information about the range of response words used for each shiritori is added to each of the shiritori response words stored in the storage means,
The output means extracts, from the storage means, response words corresponding to the range of response words used for shiritori set by the range setting means, based on information about the range of response words used for the shiritori. A voice dialogue apparatus characterized by outputting a voice. A spoken dialogue method that interacts with a speaker,
Extracting the beginning and ending from the word input via the voice input means to recognize the voice;
Based on the beginning and ending of the speech-recognized word, determine whether the input word is spoken according to the rules of shiritori,
When it is determined that the input word is uttered in accordance with the rules of shiritori, a response word starting from the same character as the end of the speech-recognized word is extracted and output as speech Spoken dialogue method.

Images