JP2015087718A - Voice interaction system and voice interaction method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To give a proper response to a long speech of a person.SOLUTION: A voice interaction system includes speech recognition means which recognizes speech uttered by a person in externally input speech, and speech output means which outputs a response speech according to a speech recognition result obtained by the speech recognition means. The speech recognition means determines whether the externally input speech is a first speech pattern determined in advance as a comma, or a second speech pattern determined in advance as a period. When the first speech pattern is determined, the speech in a comma section preceding the first speech pattern is recognized. When a second speech pattern is determined, the speech in a period section preceding the second speech pattern is recognized.

Description

  The present invention relates to a voice dialogue system and a voice dialogue method, and more particularly to a technique for recognizing a voice uttered by a person and uttering a voice corresponding to the recognized voice.

  Patent Document 1 discloses an interactive device that recognizes a user's voice and performs a dialogue with the user based on the recognized voice. As in the dialogue device disclosed in Patent Document 1, a robot that performs dialogue with a person analyzes a sentence structure in a person's utterance in the dialogue with the person, and then responds to the utterance based on the analysis result. Say the following voice.

  That is, such a robot recognizes speech in all speech sections included in one sentence in a human utterance, and then responds to the speech in the utterance. For this reason, when one sentence spoken by a person is long, there is a problem in that the response is not performed easily and the conversation becomes poor in tempo, and the conversational motivation of the person is reduced. In particular, when a person speaks unilaterally, it is difficult to understand the interruption of the speech section, and it is difficult to establish a beautiful answer alternately.

JP 2013-113966 A

  The present invention has been made on the basis of the above-described knowledge, and an object thereof is to provide a voice dialogue system and a voice dialogue method capable of appropriately receiving and answering even when a human utterance is long. To do.

  The voice dialogue system according to the first aspect of the present invention includes a voice recognition unit for recognizing a voice uttered by a person and a response voice corresponding to a voice recognition result by the voice recognition unit. Voice utterance means for uttering the voice, wherein the voice recognition means is a first voice pattern that is predetermined as a punctuation point or a predetermined first as a punctuation point. If it is determined that it is the first voice pattern, the voice in the section of the reading point unit until it is determined that the first voice pattern is determined, When it is determined that the voice pattern is the second voice pattern, the voice in the section in units of punctuation until it is determined that the voice pattern is the second voice pattern is recognized.

  The voice interaction method according to the second aspect of the present invention includes a voice recognition step for recognizing a voice uttered by a person in a voice input from the outside, and a response voice according to a voice recognition result obtained by the voice recognition step. And in the voice recognition step, the voice inputted from the outside is a first voice pattern predetermined as a punctuation mark or a predetermined first as a punctuation mark. If it is determined that it is the first voice pattern, the voice in the section of the reading point unit until it is determined that the first voice pattern is determined, When it is determined that the voice pattern is the second voice pattern, the voice in the section in units of punctuation until it is determined that the voice pattern is the second voice pattern is recognized.

  According to each aspect of the present invention described above, it is possible to provide a voice dialogue system and a voice dialogue method capable of giving an appropriate answer even when a person's utterance is long.

1 is a configuration diagram of a voice interaction system according to an embodiment. It is a flowchart which shows the process of the speech dialogue system which concerns on embodiment. It is a flowchart which shows the response sentence selection process of the speech dialogue system which concerns on embodiment. It is a flowchart which shows the process part which implement | achieves the response in the reading level which concerns on embodiment. It is a figure which shows the unit of the reading process and phrase process which concern on embodiment. It is a figure which shows the process state of the 1st step | paragraph of the 1st cycle which concerns on embodiment. It is a figure which shows the process state of the 2nd step | paragraph of the 1st cycle which concerns on embodiment. It is a figure which shows the process state of the 1st step | paragraph of the 2nd cycle which concerns on embodiment. It is a figure which shows the process state of the 2nd step | paragraph of the 2nd cycle which concerns on embodiment. It is a figure which shows the process state of the 1st step | paragraph of the 3rd cycle which concerns on embodiment. It is a figure which shows the process state of the 2nd step | paragraph of the 3rd cycle which concerns on embodiment. It is a figure which shows the process state of the 1st step | paragraph of the 4th cycle which concerns on embodiment. It is a figure which shows the process state of the 2nd step | paragraph of the 4th cycle which concerns on embodiment. It is a figure which shows the process state of the 1st step | paragraph of the 5th cycle which concerns on embodiment. It is a figure which shows the process state of the 2nd step | paragraph of the 5th cycle which concerns on embodiment. It is a flowchart which shows the process part which implement | achieves the response at the punctuation level which concerns on embodiment. It is a flowchart which shows an example of the process state of the response sentence selection process which concerns on embodiment. It is a flowchart which shows an example of the process state of the response sentence selection process which concerns on embodiment.

  Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings. Specific numerical values and the like shown in the following embodiments are merely examples for facilitating understanding of the invention, and are not limited thereto unless otherwise specified. In the following description and drawings, matters obvious to those skilled in the art are omitted or simplified as appropriate for the sake of clarity.

<Embodiment of the Invention>
With reference to FIG. 1, the structure of the voice interactive system 1 which concerns on this Embodiment is demonstrated. FIG. 1 is a configuration diagram of a voice interaction system 1 according to the present embodiment. The voice interaction system 1 can be applied to, for example, a robot that interacts with a person.

  The voice interaction system 1 includes a control unit 2, a storage unit 3, a microphone 4, a speaker 5, and an I / O port 6. The control unit 2, the storage unit 3, and the I / O port 6 are connected to each other via a bus. The microphone 4 and the speaker 5 are connected to the I / O port 6.

  The control unit 2 comprehensively controls the voice interaction system 1. The control unit 2 includes a CPU (Central Processing Unit), and executes various processes as the voice interaction system 1 according to the first embodiment by executing a program stored in the storage unit 3. That is, the program stored in the storage unit 3 includes codes for causing the CPU to execute various processes as the voice interaction system 1 according to the present embodiment. The control unit 2 includes a punctuation speech recognition unit 11, a noun extraction unit 12, a punctuation response sentence creation unit 13, an utterance timing determination unit 14, a punctuation speech recognition unit 21, a punctuation response sentence creation unit 22, and a speech synthesis unit 30. Function as.

  The reading point speech recognition unit 11 recognizes speech uttered by a person in units of reading points. The punctuation point speech recognition unit 11 utters at the punctuation point when the time during which the next utterance period does not start after the utterance period (speech period) in which the person utters the voice has reached the first predetermined time has been reached. Recognize that the section has ended. The reading point speech recognition unit 11 recognizes the speech in the section of the reading point unit including the completed utterance section. More precisely, the reading point voice recognition unit 11 recognizes the voice of the completed utterance section. That is, the reading point speech recognition unit 11 recognizes speech in one utterance section, which is a section from the start of one sentence or the previous reading point to the next reading point. Here, as the first predetermined time, a suitable time may be arbitrarily set by adopting, for example, an average time between utterance sections at a reading point in a person's utterance.

  The noun extraction unit 12 extracts nouns from the reading unit speech recognized by the reading point speech recognition unit 11 and stores the noun information indicating the extracted nouns in the noun database 42.

  The punctuation mark response sentence creation unit 13 creates a response sentence for confirming the noun indicated by the noun information stored in the noun database 42. That is, the punctuation mark response sentence generation unit 13 generates response sentence information indicating a response sentence for confirming the noun.

  The utterance timing determination unit 14 detects the end of the utterance section at the punctuation mark, and when the response sentence information is generated by the punctuation response sentence creation unit 13 at the detection timing, the utterance timing determination unit 14 converts the response sentence information into the speech synthesis unit. Output to 30. In the same manner as described above, the end of the utterance section at the punctuation point is detected when the time when the next utterance section is not started after the utterance section in which the person utters the voice has finished has reached the first predetermined time. May be detected.

  The phrase speech recognition unit 21 recognizes speech uttered by a person in phrase units. When the punctuation speech recognition unit 21 reaches a second predetermined time longer than the first predetermined time after the utterance interval in which the person utters the voice ends, the next utterance interval does not start , It recognizes that the utterance section has ended at the point. Then, the phrase speech recognition unit 21 recognizes speech in a phrase unit section including the completed speech section. More precisely, the phrase recognition voice recognition unit 21 recognizes speech in one or more utterance sections included in one sentence section up to the phrase. That is, the phrase speech recognition unit 21 recognizes speech in at least one utterance section included in the section from the start of one sentence to the phrase. Here, as the second predetermined time, for example, an appropriate time may be set by adopting, for example, an average time between utterance sections at a punctuation point in a person's utterance.

  The phrase response sentence creation unit 22 creates a response sentence for confirming the “verb” and “case element + case” recognized by the phrase recognition unit 21. That is, the phrase response sentence creation unit 22 generates response sentence information indicating a response sentence for confirming “verb” and “case element + case”, and stores the response sentence information in the speech synthesis unit 30 as an output.

  The speech synthesizer 30 generates speech information indicating speech of the response sentence indicated by the response sentence information output from the punctuation mark response sentence creation unit 13 and the phrase response sentence creation unit 22 by speech synthesis, and the I / O port 6 Is output to the speaker 5 via.

  The storage unit 3 stores the above-described program and various information necessary for processing as the voice interaction system 1. In the storage unit 3, a recognition dictionary database 41 and a noun database 42 are constructed. In the recognition dictionary database 41, speech information of a plurality of words for collation is stored in advance in order to recognize words in speech uttered by a person. The voice information of these words may be created in advance so as to indicate an average voice calculated by sampling voices of a plurality of people, for example.

  That is, the punctuation point speech recognition unit 11 and the punctuation point speech recognition unit 21 collate the speech information of words stored in the recognition dictionary database 41 with the speech information of speech uttered by the person, so that the person speaks. By recognizing each word in the voice, the voice content in the voice-recognized section is recognized. As the types of words in the recognition dictionary database 41, “verb”, “case element”, “case”, and the like are prepared.

  Here, the punctuation point speech recognition unit 11 and the punctuation point speech recognition unit 21 generate, as speech recognition results, speech contents of a plurality of patterns that are possible speech contents in the speech-recognized section. If the same “case element” (noun) is included in the voice content of the plurality of patterns in a predetermined ratio or more, the speech recognition unit for reading marks 11 reads the “case element” (noun). Is determined to be reliable. The punctuation speech recognition unit 21, when the same “verb” or “case element + case” is included in the speech content of the plurality of patterns at a predetermined ratio or more, the “verb” or “case” Judge that element + case is reliable. For example, when 10 patterns of audio content are generated, if the same “verb” is included in 8 or more patterns of audio content, it is determined that the “verb” is reliable, and the same “verb”. If the audio content that contains is less than 8 patterns, it may be determined that the “verb” is not reliable.

  The noun extraction unit 12 extracts “case elements” (nouns) determined to be reliable by the punctuation speech recognition unit 11, and the punctuation response sentence creation unit 13 extracts the “case elements” extracted by the noun extraction unit 12. Response sentence information indicating a response sentence for confirming (noun) is generated. On the other hand, the phrase response sentence creation unit 22 generates response sentence information indicating a response sentence for confirming that the “verb” and “case element + case” are determined to be reliable by the phrase recognition unit 21. To do. When a plurality of “verbs” determined to be reliable are detected, a response sentence may be created using the most reliable one among them. Here, the “verb” determined to be the most reliable may be the “verb” having the largest number of the same “verb” included in the audio contents of the plurality of generated patterns. The same applies to “case element” (noun) or “case element + case”.

  The microphone 4 converts audio input from the outside into audio information indicating the audio and outputs the audio information to the I / O port 6. The speaker 5 converts the sound information output from the I / O port 6 into sound and outputs the sound. Thereby, the voice synthesized by the voice synthesis unit 30 of the control unit 2 is uttered.

  The I / O port 6 performs A / D conversion on the audio information output from the microphone 4 and outputs it to the control unit 2. The I / O port 6 performs D / A conversion on the audio information output from the control unit 2 and outputs it to the speaker 5.

  In the present embodiment, the configuration described above allows parrots to be returned at short intervals in reading units based on the noun extraction result, and the conversation can be continued without breaking the opponent's story. And Also, according to this, even if it is difficult to make a comprehensive response to the other person's story, it is possible to return a short parrot frequently, so Can be urged.

  Then, with reference to FIGS. 2-4, the process of the voice interactive system 1 which concerns on embodiment of this invention is demonstrated. FIG. 2 is a flowchart showing processing of the voice interaction system 1 according to the embodiment of the present invention.

  The microphone 4 continuously generates sound information indicating sound input from the outside, and outputs the sound information to the control unit 2 via the I / O port 6. Therefore, when there is an utterance input from a person (S1: Yes), the microphone 4 generates voice information indicating the voice in the utterance and outputs it to the control unit 2 via the I / O port 6.

  The reading point speech recognition unit 11 detects the end of the utterance section at the reading point, and performs speech recognition of the voice indicated by the speech information in the utterance section based on the voice information in the ended utterance section (S2). Specifically, the voice recognition unit for reading marks 11 indicates that the time when the voice information indicating the voice uttered by the person from the microphone 4 is not output has reached the first predetermined time, and the end of the speech section at the reading point. Detect as. Here, the determination as to whether or not the voice information indicating the voice uttered by the person is no longer output is made based on, for example, whether or not the sound pressure level of the voice indicated by the voice information is equal to or lower than a predetermined constant value. do it. The reading voice recognition unit 11 performs voice recognition of the voice indicated by the voice information received in the utterance section.

  The noun extraction unit 12 extracts nouns determined to be reliable by the punctuation speech recognition unit 11 from the speech content recognized by the punctuation speech recognition unit 11, and stores noun information indicating the extracted nouns in the noun database 42. (S3).

  The punctuation mark response sentence generation unit 13 generates response sentence information indicating a response sentence for confirming a noun indicated by the noun information stored in the noun database 42 and stores the response sentence information in the storage unit 3 (S4).

  On the other hand, the utterance timing determination unit 14 detects the end of the utterance section at the reading (S5). Specifically, as described above, the utterance timing determination unit 14 indicates that the time when the voice information indicating the voice uttered by the person from the microphone 4 is not output has reached the first predetermined time. Detect as end.

  When the utterance timing determination unit 14 detects the end of the utterance section, the utterance timing determination unit 14 determines whether or not a response sentence for confirming the noun can be uttered (S6). Specifically, the utterance timing determination unit 14 determines that the response sentence for confirming the noun can be uttered when the response sentence information has been created by the punctuation mark response sentence creation unit 13 and stored in the storage unit 3. . On the other hand, the utterance timing determination unit 14 determines that the response sentence for confirming the noun cannot be uttered when the response sentence information has not been created by the punctuation mark response sentence creation unit 13 and is not stored in the storage unit 3.

  When the speech timing determination unit 14 determines that the response sentence for confirming the noun can be spoken (S6: Yes), the speech timing determination unit 14 outputs the response sentence information stored in the storage unit 3 to the speech synthesis unit 30. If the utterance timing determination unit 14 determines that the response sentence for confirming the noun cannot be uttered (S6: No), the response sentence information has not been created, so the response sentence information is output to the speech synthesizer 30. Absent.

  On the other hand, the speech recognition unit 21 for punctuation detects the end of the utterance section at the punctuation point, and based on the speech information in each speech section included in the one-sentence section that continues to the finished utterance section, Voice recognition indicated by the voice information is performed (S7). Specifically, the speech recognition unit 21 for a punctuation point indicates that the time when the speech information indicating the speech uttered by the person from the microphone 4 is not output has reached the second predetermined time, and the end of the utterance section at the punctuation point Detect as. Here, the determination as to whether or not the voice information indicating the voice uttered by the person is no longer output is made based on, for example, whether or not the sound pressure level of the voice indicated by the voice information is equal to or lower than a predetermined constant value. do it. Then, the phrase speech recognition unit 21 performs speech recognition of the speech indicated by the speech information received in each utterance section.

  The phrase speech recognition unit 21 determines whether or not at least one of “verb” and “case element + case” exists in the recognized speech content of one sentence (S8).

  When the phrase speech recognition unit 21 determines that at least one of “verb” and “case element + case” exists in the recognized speech content of the sentence (S8: Yes), “verb” and “case element” It is determined whether or not the “+ case” determined to exist is reliable (S9).

  The phrase response sentence creation unit 22 includes both “verb” and “case element + case” in the speech content of one sentence recognized by the phrase speech recognition unit 21, and “verb” and “case element + case”. Is determined to be reliable (S9: Yes), response sentence information indicating a response sentence confirming “verb” + “case element + case” is created and output to the speech synthesizer 30 (S10). ).

  On the other hand, the phrase response sentence creation unit 22 determines that the phrase speech recognition unit 21 determines that at least “verb” exists and only “verb” is reliable in the recognized speech content of the sentence (S11). ), Response sentence information indicating a response sentence for confirming the “verb” is created and output to the speech synthesizer 30 (S12).

  On the other hand, when the phrase response sentence creation unit 22 recognizes at least “case element + case” and only “case element + case” can be trusted in the speech content of one sentence recognized by the phrase speech recognition unit 21. If determined (S13), response sentence information indicating a response sentence for confirming “case element + case” is created and output to the speech synthesizer 30 (S14).

  When the phrase speech recognition unit 21 determines that neither “verb” nor “case element + case” exists in the recognized speech content of the sentence (S8: No), and in the recognized speech content of the sentence If all of the “verb” and “case element + case” determined to be present are determined to be unreliable (S9: No), the response sentence information is not created.

  The speech synthesizer 30 selects response text information for uttering a response text in response from the response text information output from the punctuation response text creation section 13 and the phrase response text creation section 22 (S15). The voice synthesizer 30 synthesizes the selected response sentence information, generates voice information indicating the voice of the response sentence indicated by the response sentence information, and sequentially outputs the voice information to the speaker 5 via the I / O port 6 ( S16). Thereby, as a response to the voice uttered by the person, the voice of the response sentence for confirming the content of the voice is uttered from the speaker 5.

  Here, with reference to FIG. 3, the response sentence selection process of step S15 is demonstrated in detail. FIG. 3 is a flowchart showing the response sentence selection process in step S15.

  In FIG. 3, the response sentence information created by the punctuation mark response sentence creation unit 13 is shown as “A”, and the response sentence information created by the phrase response sentence creation unit 22 is shown as “B”. When the speech synthesizer 30 receives output of response sentence information from at least one of the punctuation response sentence creation unit 13 and the phrase response sentence creation unit 22, the speech synthesis unit 30 synthesizes the response sentence indicated by the response sentence information. Start.

  Here, when the speech synthesizer 30 receives the output of the response text information B during the speech synthesis of the response text information A (during speech of the response text indicated by the response text information A), The voice synthesis of A is interrupted, and the voice synthesis of the response sentence information B is started. On the other hand, if the speech synthesizer 30 receives the output of the response text information A during the speech synthesis of the response text information B (during speech of the response text indicated by the response text information B), The speech synthesis of the information B is continued without interruption, and the speech synthesis of the response sentence information A is not performed. That is, the speech synthesizer 30 uses the response sentence indicated by the response sentence information B created by the phrase response sentence creation unit 22 rather than the response sentence shown by the response sentence information A created by the punctuation response sentence creation unit 13. Speak preferentially. Since the response sentence of the response sentence information B is a general response content for a single sentence uttered by a person, the response sentence information B has been created rather than the response sentence of the response sentence information A. This is because it is more appropriate as a response content at the timing when a person finishes speaking a sentence, as in the case of a certain timing.

  Here, when continuously synthesizing the response sentence indicated by the response sentence information A, the response sentence is replaced with a predetermined interaction (interjection) (for example, “Yes”) to perform speech synthesis. As a result, it is possible to always give a rhythm to the conversation as compared with the case of responding to confirm the noun.

  As described above, the voice conversation system 1 according to the present embodiment, as shown in FIG. 4, performs the reading level of the reading level by the process centering on the detection of the end of the utterance period (S5) and the determination of the utterance timing (S6). It is possible to respond at breaks, enabling a conversation with good tempo.

  That is, by detecting that there is no human utterance for a time (first predetermined time) shorter than the time (second predetermined time) for detecting a punctuation mark, It is possible to detect a break. As described above, this is detected when, for example, a time during which a sound pressure level of a voice uttered by a person is below a certain value continues for a first predetermined time (for example, 0.3 sec).

  When a response sentence for confirming a noun has already been created when detecting the end of the utterance section, the voice of the response sentence is uttered at the end timing of the utterance section. That is, a response sentence is uttered at the timing of a reading point in a person's utterance, and a conversation with a good tempo can be performed.

  Next, with reference to FIG. 5 to FIG. 18, the flow of processing of the above-described voice interaction system 1 will be described by taking an example of human speech.

  Here, an example in which a person speaks the following contents will be described.

“The husband goes to his relatives, but on the 11th, he said he would come home and went out in the rain.”

  In this case, as shown in FIG. 5, the speech recognition processing (reading processing) in the speech recognition unit 11 for reading marks is as follows: Say, “I went out in the rain,” and it is done in units of punctuation marks. The speech recognition process (punctuation processing) in the speech recognition unit 21 for the punctuation marks is “until the master went to his relatives, but he said he would come back on the 11th, but he went out in the rain” This is done in units of breaks at the punctuation level. In other words, while the punctuation process is performed one cycle, the reading process is performed five cycles. Hereinafter, five cycles of the reading process will be described. In the following description, the reading process will be described assuming that all nouns in the speech content of the utterance section are determined to be reliable.

(First cycle, first stage: Fig. 6)
When a person speaks up to the first utterance “the master is,” the punctuation speech recognition unit 11 and the utterance timing determination unit 14 detect the end of the utterance section in punctuation points. At this time, the utterance timing determination unit 14 detects the end of the utterance section, but has not yet created the response sentence information for the first utterance, so the utterance of the response sentence via the speech synthesizer 30 is performed. Absent. The punctuation point speech recognition unit 11 performs speech recognition of the voice “master kane” in the first utterance section.

(First cycle, second stage: Fig. 7)
The utterance timing determination unit 14 waits for the end of the second utterance section. The noun extraction unit 12 extracts the noun “master” from the speech content of the first utterance recognized by the punctuation speech recognition unit 11. Based on the noun “master” extracted by the noun extraction unit 12, the punctuation mark response sentence creation unit 13 creates response sentence information of the response sentence “Ao Master is” that confirms the noun “master”.

(Second cycle, first stage: Fig. 8)
When a person speaks up to the second utterance “I'm going to a relative”, the punctuation speech recognition unit 11 and the utterance timing determination unit 14 detect the end of the utterance section in units of punctuation. At this time, since the response sentence information for the first utterance has been created, the utterance timing determination unit 14 transmits the response sentence information to the speech synthesizer 30. The voice recognition unit for reading marks 11 performs voice recognition of the voice “I go to relatives” in the second utterance section.

(Second cycle, second stage: Fig. 9)
The utterance timing determination unit 14 waits for the end of the third utterance section. The noun extraction unit 12 extracts the noun “relative” from the speech content of the second utterance recognized by the punctuation speech recognition unit 11. Based on the noun “relative” extracted by the noun extraction unit 12, the punctuation mark response sentence creating unit 13 creates response sentence information of the response sentence “relative neighbour” that confirms the noun “relative”. In addition, the speech synthesizer 30 generates speech of the response sentence “Ao ga ga ga” indicated by the response sentence information transmitted from the utterance timing determination unit 14 by speech synthesis and utters it via the speaker 5.

(3rd cycle, 1st stage: Fig. 10)
When a person speaks up to the third utterance “day 11”, the punctuation speech recognition unit 11 and the utterance timing determination unit 14 detect the end of the utterance section in punctuation points. At this time, since the response sentence information for the second utterance has been created, the utterance timing determination unit 14 transmits the response sentence information to the speech synthesis unit 30. The voice recognition unit for reading marks 11 performs voice recognition of the voice “day 11” in the third utterance section.

  Here, the speech synthesizer 30 replaces the response sentence “11th” indicated by the response sentence information transmitted from the utterance timing determination unit 14 with the voice of the standard answer “Yes”. In this way, when the results of voice recognition processing (reading mark processing) at reading points continue, the response sentence is replaced with a simple answer (interjection) that does not check the nouns, so that the conversation has a rhythm. be able to.

(3rd cycle, 2nd stage: Fig. 11)
The utterance timing determination unit 14 waits for the end of the fourth utterance section. The noun extraction unit 12 extracts the noun “11 days” from the speech content of the third utterance recognized by the punctuation speech recognition unit 11. Based on the noun “11 days” extracted by the noun extraction unit 12, the punctuation mark response sentence creating unit 13 creates response sentence information of the response sentence “11 days” that confirms the noun “11 days”. Further, the speech synthesizer 30 generates speech of the response sentence “Yes” after replacement by speech synthesis and utters it through the speaker 5. As described above, in the response sentence selection process in the speech synthesizer 30, when the result of the speech recognition process (reading mark process) at the reading points continues, the “response for confirming the noun” and the standard answer “Yes” are alternately spoken. Synthesize.

  Here, for example, the standard interaction is stored in advance in the storage unit 3 as response sentence information indicating the response sentence of the interaction, and the speech synthesizer 30 reads the speech of the response sentence (consideration) indicated by the response sentence information. May be synthesized with speech. Also, the contents of the considerations are not limited to the above example “Yes”, but other considerations may be prepared, and a plurality of considerations may be prepared and synthesized in a predetermined order or randomly. Good.

(4th cycle, 1st stage: Fig. 12)
When a person speaks up to the fourth utterance “Please say come back”, the punctuation speech recognition unit 11 and the utterance timing determination unit 14 detect the end of the utterance section in units of punctuation. At this time, since the response sentence information for the third utterance has been created, the utterance timing determination unit 14 transmits the response sentence information to the speech synthesis unit 30. The punctuation point speech recognition unit 11 performs speech recognition of the speech “Please come home” in the fourth utterance section.

(4th cycle, 2nd stage: Fig. 13)
The utterance timing determination unit 14 waits for the end of the fifth utterance section. The noun extraction unit 12 cannot extract a noun because the speech content of the fourth utterance recognized by the punctuation point speech recognition unit 11 does not include a noun. Therefore, the punctuation mark response sentence creation unit 13 does not create response sentence information for the fourth utterance. In addition, the speech synthesizer 30 generates speech of the response sentence “Nine days” indicated by the response sentence information output from the utterance timing determination unit 14 by speech synthesis, and utters it through the speaker 5.

(5th cycle, 1st stage: Fig. 14)
When a person speaks up to the fifth utterance “I went out in the rain”, the punctuation speech recognition unit 11 and the utterance timing determination unit 14 detect the end of the utterance section in units of punctuation. At this time, although the end of the utterance period is detected, the utterance timing determination unit 14 does not utter the response sentence via the speech synthesizer 30 because the response sentence information for the fourth utterance has not been created. . The voice recognition unit for reading marks 11 performs voice recognition of the voice “I went out in the rain” in the fifth utterance section.

(5th cycle, 2nd stage: Fig. 15)
The utterance timing determination unit 14 waits for the end of the sixth utterance section. The noun extraction unit 12 extracts the noun “rain” from the speech content of the fifth utterance recognized by the punctuation speech recognition unit 11. Based on the noun “rain” extracted by the noun extraction unit 12, the punctuation mark response sentence creation unit 13 creates response sentence information of the response sentence “rain” that confirms the noun “rain”.

  On the other hand, in the fifth cycle, one sentence from a person said, “My husband goes to his relatives, but he says he will come home on the 11th.” Since the utterance ends, the phrase speech recognition unit 21 also detects the end of the utterance section at the phrase. Therefore, as shown in FIG. 16, the punctuation speech recognition unit 21 also performs speech recognition processing (punctuation processing) for one sentence of speech from the first utterance to the fifth utterance.

  Here, it is assumed that the phrase recognition unit 21 determines that the reliable “case element + case” is “the master” and that the reliable verb is “going out”. In this case, the response sentence creation unit 22 for punctuation generates response sentence information of a response sentence “My husband went out” which is the most reliable combination of “verb” and “case element + case”. Then, the phrase response sentence creation unit 22 transmits the generated response sentence information to the speech synthesis unit 30.

  The response sentence creation unit 22 for punctuation generates response sentence information of a response sentence “I went out” when only “verb” is reliable, and when only “case element + case” is reliable. , Response sentence information of the sentence “My husband is.” Is generated. When the “verb” and “case element + case” are unreliable, the phrase response sentence creation unit 22 does not generate response sentence information.

(6th cycle)
In the sixth cycle, depending on the response sentence information creation / transmission timing of the punctuation response sentence creation unit 13 and the phrase response sentence creation unit 22, the speech synthesizer 30 causes the response sentence “rain” The speech synthesis of “Ne” and the speech synthesis of the response sentence “My husband went out” of the phrase response sentence creation unit 22 compete.

  As described above, when only one of the response sentence information is received, the voice synthesis unit 30 synthesizes the voice of the response sentence of the response sentence information, and both response sentence information competes. In this case, the response text information from the phrase response text creation unit 22 is preferentially processed. That is, for example, as illustrated in FIG. 17, when only the response sentence information is received from the punctuation mark response sentence creation unit 13, the speech synthesizer 30 determines that the response sentence “rain” is indicated by the response sentence information. Perform voice synthesis.

  On the other hand, for example, as illustrated in FIG. 18, the response sentence information only from the punctuation mark response sentence creation unit 13 is received first, and the response sentence “rain rain” indicated by the response sentence information is being synthesized. When the response sentence information is transmitted from the phrase response sentence creation unit 22, the speech synthesis unit 30 synthesizes the response sentence “rain” indicated by the response sentence information from the punctuation response sentence creation unit 13. , And the speech synthesis of the response sentence “The master went out” indicated by the response sentence information from the phrase response sentence creation unit 22 is started. Therefore, for example, even if the voice is being spoken such as “Ama, the husband has gone out”, the voice of the response sentence having a high priority is forcibly switched.

  In summary, the dialogue result of the example sentence is as follows. “A” indicates the utterance content of the person, and “B” indicates the utterance content of the voice dialogue system 1.

A: My husband (noun extraction: my husband) goes to my relatives,
B: Oh my husband (reading result)
A: On the 11th (noun extraction: 11th)
B: Yeah (reading result)
A: Tell me to come home,
B: 11 days (reading result)
A: I went out in the rain. . . (Noun extraction: rain)
B: Your husband went out (results of the phrase processing)
A: Then suddenly ...

  As described above, in this embodiment, it is possible to improve the continuity of the dialogue by frequently responding and indicating that the voice dialogue system 1 is listening to the person. In other words, by reading frequently with short response sentences, punctuation can be encouraged so as not to disturb the person. On the other hand, by confirming with a general response sentence longer than that at the point, a person can obtain a sufficient sense of dialogue as a speaker.

  On the other hand, when the present embodiment is not applied, the dialogue result of the example sentence is as follows.

A: My husband went to my relatives, but on the 11th, he said he would come home and went out in the rain. . .
B: My husband went out (My husband went out, etc.)

  That is, no response is made at all until the utterance of one sentence up to a punctuation point ends, and the person's willingness to conversation is reduced.

  As described above, the present embodiment is a voice recognition means for recognizing a voice in a voice section in which a person utters a voice (corresponding to the reading voice recognition unit 11 and the punctuation voice recognition unit 21). , Voice utterance means for uttering the response voice according to the voice recognition result by the voice recognition means (noun extraction unit 12, punctuation response sentence creation part 13, utterance timing judgment part 14, phrase response sentence creation part 22, speech synthesis part 30). Then, the voice recognition means recognizes the voice in the section of the reading point unit including the ended voice section when the time when the next voice section does not start after the end of the voice section has reached the first predetermined time, When the time when the next voice segment is not started after the end of the voice segment has reached a second predetermined time longer than the first predetermined time, the voice in the phrase unit segment including the completed voice segment is recognized. I am doing so.

  According to this, it is possible to distinguish and recognize the punctuation mark and the punctuation mark according to the time when the next voicing period is started after the end of the speech section, and to receive and answer each punctuation mark during the utterance of the person. Therefore, even if a person's utterance is long, it is possible to give an appropriate answer.

  Further, in the present embodiment, when the utterance timing of the response voice according to the speech recognition result in units of reading points competes with the utterance timing of the response speech according to the speech recognition result in units of punctuation points, the speech recognition result in units of punctuation points The response voice corresponding to the voice is preferentially uttered. According to this, an appropriate voice can be uttered according to the punctuation marks.

<Other embodiments of the present invention>
In the above embodiment, reading and punctuation are determined based on whether or not the time when there is no utterance (the time during which the sound pressure level of the voice is below a certain value) has reached a predetermined time. I can't. For example, as described below (Modification 1) or (Modification 2), reading marks and punctuation marks may be determined.

(Modification 1: Discrimination by frequency)
When the ratio of the spectrum included in the specific frequency band is equal to or less than the first ratio, it is determined as a reading point, and when the ratio of the spectrum included in the specific frequency band is equal to or less than the second ratio lower than the first ratio May be determined to be a punctuation mark. For example, when the ratio of the spectrum included in the specific frequency band is 20% or less of the whole, it is determined as a reading point, and when the ratio of the spectrum included in the specific frequency band is 10% or less, it is determined as a punctuation mark.

  Specifically, in the voice indicated by the voice information output from the microphone 4, the percentage of the spectrum included in the specific frequency band is equal to or less than the first percentage in the reading point voice recognition unit 11 and the speech timing determination unit 14. Is detected as the end of the utterance section at the reading point. In addition, the speech recognition unit 21 for punctuation indicates that in the speech indicated by the speech information output from the microphone 4, the proportion of the spectrum included in the specific frequency band is equal to or less than the second proportion, and the utterance interval at the punctuation is ended. Detect as.

  The specific frequency band may be determined in advance as an arbitrary frequency band that is considered to have a high spectrum ratio during human speech.

(Modification 2: Discrimination based on HMM (Hidden Markov Model))
HMM for reading recognition that calculates the likelihood (reading likelihood) of a reading point based on speech, HMM for phrase recognition that calculates the likelihood (punctuation likelihood) of phrase based on speech, and the likelihood of speech (utterance) Various HMMs such as an utterance recognition HMM for calculating the section likelihood) are prepared in advance. When the likelihood calculated by the HMM for reading point recognition based on the voice is the highest, it is determined that the reading point is read. When the likelihood calculated by the HMM for recognizing a punctuation based on speech is the highest, it is determined that the punctuation is a punctuation. When the likelihood calculated by the speech recognition HMM based on the speech is the highest, it is determined that the speech is not a punctuation mark nor a punctuation mark.

  Specifically, an HMM for reading point recognition, an HMM for phrase recognition, and an HMM for speech recognition are generated in advance by learning, and information on these HMMs is stored in the storage unit 3. The punctuation point speech recognition unit 11 and the utterance timing determination unit 14 use the speech indicated by the speech information output from the microphone 4 as an input to each HMM, and the HMM for reading point recognition uses the likelihood output by those HMMs. It is detected that the calculated likelihood is the highest as the end of the utterance section at the reading point. In addition, the speech recognition unit 21 for punctuation uses the speech indicated by the speech information output from the microphone 4 as input to each HMM, and the likelihood calculated by the HMM for punctuation recognition in the likelihood output by those HMMs Is detected as the end of the utterance section at the punctuation point.

  As described above in the embodiment and the other embodiments, in the present embodiment, the first voice pattern (the next voice section is determined in advance) that the voice uttered by a person is a reading point. When it is determined that the first predetermined time has elapsed, the spectrum ratio is equal to or less than the first ratio, and the likelihood calculated by the HMM for reading mark recognition is the highest) Recognize the voice in the section of the reading point until it is determined that there is a second voice pattern that has been pre-determined that the voice uttered by the person is the reading point (the time when the next voice section is not started) 2 is determined to be the second speech pattern when it is determined that the predetermined time elapses in 2 and the spectrum ratio is equal to or less than the second ratio and the HMM for the phrase recognition is the highest likelihood). Up to the reading unit That is to recognize the voice.

  According to this, it is possible to distinguish and recognize a punctuation mark and a reading mark depending on whether it is the first sound pattern or the second sound pattern, and receive and answer each reading mark during a person's utterance. Therefore, even if a person's utterance is long, it is possible to give an appropriate answer.

  Note that the present invention is not limited to the above-described embodiment, and can be changed as appropriate without departing from the spirit of the present invention.

DESCRIPTION OF SYMBOLS 1 Voice dialogue system 2 Control part 3 Memory | storage part 4 Microphone 5 Speaker 6 I / O port 11 Speech recognition part 12 for punctuation Noun extraction part 13 Response sentence preparation part 14 for punctuation Timing determination part 21 Speech recognition part 22 for punctuation Response sentence creation unit 30 Speech synthesis unit

Claims (7)

Speech recognition means for recognizing speech uttered by a person in speech input from the outside;
Voice utterance means for uttering a response voice according to a voice recognition result by the voice recognition means,
The voice recognition means
Determining whether the externally input voice is a first voice pattern predetermined as a punctuation mark or a second voice pattern predetermined as a punctuation mark;
If it is determined to be the first sound pattern, it recognizes the sound in the reading unit interval until it is determined to be the first sound pattern;
If it is determined to be the second voice pattern, it recognizes the voice in the period of the phrase unit until it is determined to be the second voice pattern.
Spoken dialogue system. The voice recognition means
Determine the voice segment in which the person is speaking,
As the first voice pattern, it is determined that a time when the next voice section is not started after the end of the voice section has reached a first predetermined time,
Determining, as the second voice pattern, that a time during which the next voice section is not started after the end of the voice section has reached a second predetermined time longer than the first predetermined time;
The speech dialogue system according to claim 1. The speech utterance means, when the utterance timing of the response speech according to the speech recognition result in the reading unit and the utterance timing of the response speech in accordance with the speech recognition result in the phrase unit compete, Preferentially utter response voice according to the result,
The voice interaction system according to claim 1 or 2. The voice utterance unit utters a voice of a response sentence for confirming a voice content in the section of the reading point unit as a response voice according to the voice recognition result of the reading point unit,
The voice utterance unit utters a compatible voice instead of the voice of the response sentence as the response voice when the utterance of the response voice according to the voice recognition result of the reading mark unit is continuous.
The voice interaction system according to any one of claims 1 to 3. The voice utterance means has a noun extraction means for extracting a noun included in the voice content in the section of the reading point unit based on the voice recognition result of the reading point unit.
The voice utterance means utters the voice of a response sentence for confirming the noun extracted by the noun extraction means as the response voice according to the voice recognition result in reading points.
The spoken dialogue system according to any one of claims 1 to 4. The voice utterance means has a response sentence creation means for creating a response sentence for confirming the voice content in the section of the reading point unit as the content of the response voice according to the voice recognition result of the reading point unit,
The voice utterance means includes
If a response sentence is created by the response sentence creation means when it is determined that the first voice pattern is the response, the created response is used as a response voice according to the speech recognition result in the reading unit. Speak the sentence,
In the case where a response sentence has not been created by the response sentence creating means, if the response sentence has been created by the response sentence creating means when it is determined that it is the first voice pattern next time, the reading unit Utter the voice of the created response sentence as a response voice according to the voice recognition result of
The voice interaction system according to any one of claims 1 to 5. A speech recognition step for recognizing speech uttered by a person in speech input from outside;
A voice utterance step of uttering a response voice according to the voice recognition result of the voice recognition step,
In the voice recognition step,
Determining whether the externally input voice is a first voice pattern predetermined as a punctuation mark or a second voice pattern predetermined as a punctuation mark;
If it is determined to be the first sound pattern, it recognizes the sound in the reading unit interval until it is determined to be the first sound pattern;
If it is determined to be the second voice pattern, it recognizes the voice in the period of the phrase unit until it is determined to be the second voice pattern.
Spoken dialogue method.

Images