JP2017156688A - Conversation evaluation device and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a conversation evaluation device which objectively evaluates conversations by sounds.SOLUTION: A conversation evaluation device 100 includes: a sound collector 22 for generating a sound signal X1 representing a sound V1 produced by a user U1; a sound collector 24 for generating a sound signal X2 representing a sound V2 produced by a user U2; a feature acquisition unit 32 for acquiring the feature amount of the sounds forming a conversation; an information generation unit 34 for generating information R related to the conversation other than the feature amount; and a conversation evaluation unit 36 for evaluating the conversation based on the feature amount and the related information R.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a technique for evaluating a voice conversation.

  Conventionally, techniques for evaluating various matters by analyzing uttered speech have been proposed. For example, Patent Document 1 discloses a technique for estimating a speaker's psychological or physiological state from the interval of fundamental sounds in a pitch sequence of speech sound.

Japanese Patent No. 4495907

  However, in the technique of Patent Document 1, only the state of a specific speaker is estimated, and for example, a speech conversation between a plurality of speakers (for example, an impression of a sound of a response to an utterance) is objectively evaluated. I can't. In view of the above circumstances, an object of the present invention is to objectively evaluate a voice conversation.

  In order to solve the above problems, a conversation evaluation apparatus according to the first aspect of the present invention generates a feature acquisition unit that acquires a feature amount of speech constituting a conversation, and generates related information different from the feature amount for the conversation. And a conversation evaluation unit that evaluates conversation according to the feature amount and the related information. In the above aspect, it is possible to objectively evaluate the conversation according to the feature amount of the voice constituting the conversation. In addition, since related information different from the feature quantity is added to the conversation evaluation together with the feature quantity, it is possible to appropriately evaluate the conversation in comparison with a configuration in which only the feature quantity is reflected in the evaluation.

  In a preferred aspect of the present invention, the feature acquisition unit acquires a feature amount under conditions according to the related information. In the above aspect, since the feature amount is acquired under conditions according to the related information, there is an advantage that the feature amount can be appropriately acquired as compared with a configuration in which the related information is not used for acquiring the feature amount.

  A conversation evaluation device according to a second aspect of the present invention includes a feature acquisition unit that acquires a feature amount of speech that constitutes a conversation, an information generation unit that generates related information different from the feature amount for conversation, and a feature amount And a conversation evaluation unit that evaluates the conversation in response, and the feature acquisition unit acquires the feature amount under conditions according to the related information. In the above aspect, it is possible to objectively evaluate the conversation according to the feature amount of the voice constituting the conversation. In addition, since the feature amount is acquired under conditions according to the related information, there is an advantage that the feature amount can be appropriately acquired as compared with a configuration in which the related information is not used for acquiring the feature amount.

  In a preferred example of the conversation evaluation device according to each aspect described above, the feature acquisition unit acquires the pitch of each of the first voice and the second voice constituting the conversation as a feature amount, and the conversation evaluation unit receives the first voice. The conversation is evaluated according to the pitch difference between the voice and the second voice. In the above aspect, since the conversation is evaluated according to the pitch difference between the first voice and the second voice, the quality of the response voice is determined from the viewpoint of the relationship between the pitch of the response voice and the pitch of the utterance voice. It is possible to evaluate objectively.

  In each of the above-described aspects, the related information includes, for example, at least a conversational time situation, a history of past conversations between conversational speakers, a relationship between conversational speakers, and an attribute of each conversational speaker. This is one piece of information.

It is a block diagram of the conversation evaluation apparatus of 1st Embodiment. It is a flowchart of a conversation evaluation process. It is a block diagram of the conversation evaluation apparatus of 2nd Embodiment. It is a block diagram of the conversation evaluation apparatus of 3rd Embodiment. It is a block diagram of the conversation evaluation apparatus of 4th Embodiment.

<First Embodiment>
FIG. 1 is a configuration diagram of a conversation evaluation apparatus 100 according to the first embodiment of the present invention. The conversation evaluation apparatus 100 according to the first embodiment is an analysis apparatus that evaluates a conversation between the user U1 and the user U2, and is suitably used for training a good impression, for example. A conversation is composed of the voice V1 (example of the first voice) that the user U1 pronounces and the voice V2 (example of the second voice) that the user U2 pronounces.

  In the first embodiment, for example, the voice U1 of an utterance including a question and a talk is pronounced by the user U1, and a voice V2 of a response including an answer to the question from the user U1 and a response to the talk is received by the user. Assume that U2 is pronounced. The voice V2 generated by the user U2 is a voice meaning an interjection, for example. For example, “Yes”, “Yes”, etc., “E-to”, “Ano ~” etc. (stagnation of response), “Yes”, “No” answer (affirmative / negative) Phrases such as “Oh” and “O” representing the impression of the speaker, or phrases such as “E?” And “What?” That mean a question answer (re-listening) to the utterance may be exemplified as interjections.

  As illustrated in FIG. 1, the conversation evaluation apparatus 100 according to the first embodiment includes a control device 12, a storage device 14, a display device 16, an input device 18, a sound collection device 22, and a sound collection device 24. It is realized with. For example, the conversation evaluation apparatus 100 can be realized by a portable information processing apparatus such as a mobile phone or a smartphone or an information processing apparatus such as a personal computer. Note that the conversation evaluation apparatus 100 can be realized by a plurality of apparatuses configured separately from each other.

  The sound collecting device 22 and the sound collecting device 24 are sound input devices that pick up surrounding sounds. The sound collecting device 22 generates a sound signal X1 representing the sound V1 sounded by the user U1, and the sound collecting device 24 generates a sound signal X2 representing the sound V2 sounded by the user U2. The A / D converter for converting each of the audio signal X1 and the audio signal X2 from analog to digital is not shown for convenience.

  The control device 12 includes a processing circuit such as a CPU (Central Processing Unit), for example, and comprehensively controls each element of the conversation evaluation device 100. Specifically, the control device 12 evaluates the conversation between the user U1 and the user U2 by analyzing the audio signal X1 generated by the sound collection device 22 and the audio signal X2 generated by the sound collection device 24. To do. The control device 12 according to the first embodiment calculates an index (hereinafter referred to as “evaluation value”) S of impression quality for the response of the user U2 to the utterance of the user U1.

  The display device 16 (for example, a liquid crystal display panel) displays various images under the control of the control device 12. For example, the evaluation result (evaluation value S) of the conversation between the user U1 and the user U2 is displayed on the display device 16. The input device 18 receives an instruction from the user U (for example, the user U1 or the user U2) to the conversation evaluation device 100. For example, a plurality of operators operated by the user U (U 1, U 2) and a touch panel for detecting contact with the display surface of the display device 16 are preferably used as the input device 18.

  The storage device 14 stores a program executed by the control device 12 and various data used by the control device 12. For example, a known recording medium such as a semiconductor recording medium or a magnetic recording medium, or a combination of a plurality of recording media can be arbitrarily employed as the storage device 14. The control device 12 according to the first embodiment executes a program stored in the storage device 14 to thereby execute a plurality of functions (feature acquisition unit 32, information generation for evaluating conversation between the user U1 and the user U2). Unit 34 and conversation evaluation unit 36). A configuration in which the functions of the control device 12 are distributed to a plurality of devices, or a configuration in which a dedicated electronic circuit realizes part or all of the functions of the control device 12 may be employed.

  The feature acquisition unit 32 acquires the feature amount of the voice V1 of the user U1 and the feature amount of the voice V2 of the user U2. The feature acquisition unit 32 of the first embodiment extracts the feature amount of the voice V1 of the user U1 by analyzing the voice signal X1, and extracts the feature amount of the voice V2 of the user U2 by analyzing the voice signal X2. Specifically, feature values related to prosody are extracted for each of the voice V1 and the voice V2. Prosody is a linguistic and phonetic characteristic that can be perceived by the listener, and means a property that cannot be grasped only by the general notation of the language.

  The feature acquisition unit 32 of the first embodiment extracts the pitch P1 of the voice V1 of the user U1 and the pitch P2 of the voice V2 of the user U2 as feature amounts. For example, the feature acquisition unit 32 extracts an average pitch P1 in the speech section of the speech signal X1 and an average pitch P2 in the speech section of the speech signal X2. The utterance section is a section where the utterance continues (a section from the start point to the end point of a series of utterances). For the extraction of the pitch P1 and the pitch P2, a known voice analysis technique can be arbitrarily adopted.

  When the user U2 responds with the voice V2 of the pitch P2 having a specific relationship with the pitch P1 of the voice V1 uttered by the user U1, the user U1 is comfortable with the voice V2 of the user U2. There is a tendency to perceive it as a good impression with a sense of security. Specifically, when the user U2 pronounces the pitch P2 having a cooperative relationship with the pitch P1 of the user U1, the response of the user U2 is felt as a good impression. In addition, what particularly affects the impression of the response by the user U2 is the end section close to the start point of the speech section of the voice V2 among the speech sections of the voice V1 by the user U1. Therefore, the feature acquisition unit 32 of the first embodiment identifies the pitch P1 of a predetermined length (for example, 180 msec) section located at the end of the utterance section of the utterance section of the voice V1 of the user U1.

  The information generation unit 34 in FIG. 1 generates information (hereinafter referred to as “related information”) R different from the feature amount extracted by the feature acquisition unit 32 for the conversation between the user U1 and the user U2. The related information R is information related to the conversation between the user U1 and the user U2. In 1st Embodiment, the relevant information R which shows the time condition of conversation is illustrated. Specifically, the information generation unit 34 generates related information R indicating the conversation date and time (for example, date and time) and the conversation duration (for example, elapsed time from the start of conversation) as the conversation status. For example, the information generation unit 34 specifies the conversation date and time and the conversation continuation length with reference to the time counted by a timing circuit (not shown). That is, for example, the start date and time of the latest utterance section in the voice V1 or voice V2 is specified as the conversation date and time, and the elapsed time from the start time of the earliest utterance section in the voice V1 or voice V2 to the current time is set as the conversation continuation length. Identified.

  The conversation evaluation unit 36 performs a conversation between the user U1 and the user U2 according to the feature amount (pitch P1, pitch P2) extracted by the feature acquisition unit 32 and the related information R generated by the information generation unit 34. evaluate. That is, the conversation evaluation unit 36 calculates the evaluation value S according to the pitch P1, the pitch P2, and the related information R. As understood from the above description, in the first embodiment, not only the feature amount of the voice (V1, V2) constituting the conversation but also related information R other than the feature amount is considered in the conversation evaluation. The evaluation value S calculated by the conversation evaluation unit 36 is displayed on the display device 16.

  As described above, when the user U2 responds with the voice V2 of the pitch P2 having a cooperative relationship with the pitch P1 of the voice V1 of the user U1, the user U1 tends to feel a good impression. is there. Considering the above tendency, the conversation evaluation unit 36 of the first embodiment calculates the evaluation value S according to the pitch difference ΔP (ΔP = | P1−P2 |) between the pitch P1 and the pitch P2. . Specifically, the conversation evaluation unit 36 calculates the evaluation value S so that the evaluation value S becomes larger as the pitch difference ΔP is closer to the cooperative relationship. The cooperative relationship exemplified in the first embodiment is, for example, a pitch relationship in which the frequency ratio is close to an integer ratio (for example, complete once, complete eighth, complete fifth, and complete fourth).

  As described above, the conversation evaluation unit 36 of the first embodiment reflects the pitch P1 and the pitch P2 in the evaluation value S, and also includes related information R related to the conversation between the user U1 and the user U2. Then, the evaluation value S is calculated. A specific relationship between the related information R (conversation date and time, conversation continuation length) and the evaluation value S is illustrated below.

  Since conversations at night and on holidays are likely to be conversations between close friends, the user U1 is the user compared to, for example, weekday daytime conversations (typically business conversations, etc.). There is a tendency that U2 has a high possibility of taking a good impression. Considering the above tendency, when the conversation date and time specified by the related information R corresponds to nighttime or holiday, the evaluation value S is larger than the case where the conversation date and time corresponds to daytime on weekdays. Thus, the conversation evaluation unit 36 calculates an evaluation value S. For example, the conversation evaluation unit 36 adds a predetermined value to the evaluation value S when the conversation date / time corresponds to nighttime or a holiday.

  In addition, when the conversation continues for a long time, it is highly likely that the conversation between the user U1 and the user U2 is alive while feeling a good impression of each other. Considering the above tendency, the conversation evaluation unit 36 calculates the evaluation value S so that the evaluation value S becomes larger as the conversation duration length specified by the related information R is longer. For example, the conversation evaluation unit 36 adds a predetermined value to the evaluation value S when the conversation continuation length exceeds a predetermined threshold. On the other hand, when the conversation is excessively long, the impression between the users U1 and U2 may deteriorate due to fatigue of the users U1 and U2. Considering the above tendency, the evaluation value S can be deducted when the conversation duration specified by the related information R exceeds a predetermined threshold. A plurality of different threshold values can be used. For example, the first threshold value and the second threshold value are set (first threshold value <second threshold value), and the evaluation value S is added when the conversation duration is a numerical value between the first threshold value and the second threshold value. When the conversation continuation length is lower than the first threshold value or higher than the second threshold value, a configuration in which the evaluation value S is deducted is assumed. It is also possible to change the added point value or the deducted value for the evaluation value S step by step for each range having each threshold as a boundary.

  FIG. 2 is a flowchart of processing (hereinafter referred to as “conversation evaluation processing”) in which the control device 12 of the first embodiment evaluates the conversation between the user U1 and the user U2. For example, the conversation evaluation process is started with an instruction from the user U (U1, U2) to the input device 18 or the start of utterance by the user U.

  When the conversation evaluation process of FIG. 2 is started, the feature acquisition unit 32 sequentially analyzes the pitch P1 of the voice V1 of the user U1 and the pitch P2 of the voice V2 of the user U2 by analyzing the voice signal X1 and the voice signal X2. (SA1). In addition, the information generation unit 34 generates the related information R (the conversation date and time and the conversation continuation length in the first embodiment) with reference to the time counted by the timing circuit, for example (SA2). The conversation evaluation unit 36 calculates an evaluation value S according to the feature amount (pitch P1, pitch P2) extracted by the feature acquisition unit 32 and the related information R generated by the information generation unit 34 (SA3). The feature extraction by the feature acquisition unit 32 (SA1) and the generation of the related information R by the information generation unit 34 (SA2) can be reversed.

  As illustrated above, in the first embodiment, it is possible to objectively evaluate the conversation between the user U1 and the user U2 according to the feature amounts of the voice V1 and the voice V2 constituting the conversation. . Further, since the conversation is evaluated in consideration of the related information R of the conversation in addition to the feature amounts of the voice V1 and the voice V2 constituting the conversation, the conversation is compared with the configuration in which only the feature amount is reflected in the evaluation result. Appropriate evaluation is possible. In the first embodiment, in particular, since the conversation is evaluated according to the pitch difference ΔP between the pitch P1 of the voice V1 and the pitch P2 of the voice V2, the voice V2 of the user U2 with respect to the voice V1 of the user U1. From the viewpoint of the pitch (ie, pitch difference), it is possible to objectively evaluate the quality of the impression of the response by the user U2.

Second Embodiment
A second embodiment of the present invention will be described. In addition, about the element which an effect | action and function are the same as that of 1st Embodiment in each form illustrated below, the reference | standard referred by description of 1st Embodiment is diverted, and each detailed description is abbreviate | omitted suitably.

  FIG. 3 is a configuration diagram of the conversation evaluation apparatus 100 according to the second embodiment. As illustrated in FIG. 3, the storage device 14 of the conversation evaluation device 100 according to the second embodiment stores history information H for each combination of users U. The history information H is information relating to the history of conversations between the users U (conversation history). Specifically, the history information H of the second embodiment includes the frequency of conversations between users U in the past (hereinafter referred to as “conversation frequency”) and the elapsed time from the first conversation between users U. (Hereinafter referred to as “related period”). The conversation frequency means the number of conversations within a predetermined length period (for example, one month). The conversation frequency of the related information R is updated for each conversation between the users U. The conversation frequency and the relationship period can also be restated as an index of intimacy between users U.

  The information generation unit 34 of the second embodiment generates the related information R with reference to the history information H stored in the storage device 14. For example, the user U1 and the user U2 operate the input device 18 as appropriate to instruct the conversation evaluation device 100 about their identification information. The information generation unit 34 searches the storage device 14 for history information H between the user U1 and the user U2 indicated by the identification information, and related information R including the conversation frequency and the related period specified by the history information H. Is generated. Note that the operation of the feature acquisition unit 32 extracting the pitch P1 of the voice V1 and the pitch P2 of the voice V2 is the same as in the first embodiment.

  When the conversation frequency is high, the user U1 and the user U2 are in a good relationship, and there is a high possibility that the user U1 and the user U2 are talking while maintaining a good impression. Considering the above tendency, the conversation evaluation unit 36 calculates the evaluation value S so that the higher the conversation frequency specified by the related information R is, the larger the evaluation value S becomes. For example, the conversation evaluation unit 36 adds a predetermined value to the evaluation value S when the conversation frequency exceeds a predetermined threshold. A plurality of different threshold values can be used. For example, when the conversation frequency is a numerical value between the first threshold value and the second threshold value (first threshold value <second threshold value), the evaluation value S is added, while the conversation frequency falls below the first threshold value or the second When exceeding a threshold value, the structure which deducts evaluation value S is assumed. It is also possible to change the added point value or the deducted value for the evaluation value S step by step for each range having each threshold as a boundary.

  Similarly, when the relationship period from the first conversation is long, there is a high possibility that the user U1 and the user U2 have a good relationship and are talking while maintaining a good impression. In consideration of the above tendency, the conversation evaluation unit 36 calculates the evaluation value S so that the longer the relation period specified by the related information R, the larger the evaluation value S. For example, the conversation evaluation unit 36 adds a predetermined value to the evaluation value S when the relationship period exceeds a predetermined threshold. A plurality of different threshold values can be used. For example, when the relationship period is a numerical value between the first threshold value and the second threshold value (first threshold value <second threshold value), the evaluation value S is added, while the relationship period falls below the first threshold value or the second When exceeding a threshold value, the structure which deducts evaluation value S is assumed. It is also possible to change the added point value or the deducted value for the evaluation value S step by step for each range having each threshold as a boundary.

  As exemplified above, also in the second embodiment, the conversation is evaluated in consideration of the related information R of the conversation in addition to the feature amounts of the voice V1 and the voice V2 constituting the conversation. Therefore, similarly to the first embodiment, it is possible to appropriately evaluate the conversation as compared with the configuration in which only the feature amount is reflected in the evaluation. Particularly in the second embodiment, since the conversation history (for example, conversation frequency and relationship period) between the users U is used as the related information R, appropriate evaluation based on the tendency of past conversations between the users U is realized. Is done.

  In the above description, the identification information instructed by the operation on the input device 18 is used, but the method for specifying the user U1 and the user U2 is arbitrary. For example, the user U1 is specified by speaker identification with respect to the audio signal X1 and the user U2 is specified by speaker identification with respect to the audio signal X2, and the history information H between the user U1 and the user U2 is searched. Is also possible. A known recognition technique can be arbitrarily adopted for speaker identification of the user U1 and the user U2.

<Third Embodiment>
FIG. 4 is a configuration diagram of the conversation evaluation apparatus 100 of the third embodiment. As illustrated in FIG. 4, the storage device 14 of the conversation evaluation device 100 according to the third embodiment stores speaker information Q for each combination of users U. The speaker information Q is information indicating the relationship between the users U. Specifically, the speaker information Q of the third embodiment specifies the mutual relationship (friend, family, acquaintance, colleague, etc.) between the users U and the closeness between the users U. The interrelationship and closeness can be specifically instructed by an operation from the user U with respect to the input device 18, but can be reflected in the speaker information Q from information registered in SNS (Social Networking Service), for example. It is.

  The information generation unit 34 of the third embodiment generates related information R with reference to the speaker information Q stored in the storage device 14. For example, the information generation unit 34 searches the storage device 14 for speaker information Q between the user U1 and the user U2 specified by the input of identification information or speaker identification as in the second embodiment, and The related information R including the mutual relationship and the intimacy specified by the speaker information Q is generated. Note that the operation of the feature acquisition unit 32 extracting the pitch P1 of the voice V1 and the pitch P2 of the voice V2 is the same as in the first embodiment.

  When the mutual relationship is a friend, there is a high possibility that the user U1 and the user U2 have a good relationship and are talking while maintaining a good impression. In consideration of the above tendency, when the mutual relationship specified by the related information R is a friend, the conversation value is set so that the evaluation value S is larger than that when the mutual relationship is another relationship. The evaluation unit 36 calculates an evaluation value S. Specifically, the conversation evaluation unit 36 adds a predetermined value to the evaluation value S when the mutual relationship between the user U1 and the user U2 is a friend.

  Similarly, when the intimacy is high, the user U1 and the user U2 are in a good relationship, and there is a high possibility that they are talking while maintaining a good impression. In consideration of the above tendency, the conversation evaluation unit 36 calculates the evaluation value S so that the higher the familiarity specified by the related information R, the higher the evaluation value S. For example, the conversation evaluation unit 36 adds a predetermined value to the evaluation value S when the familiarity exceeds a predetermined threshold. A plurality of different threshold values can be used. For example, when the familiarity is a numerical value between the first threshold value and the second threshold value (first threshold value <second threshold value), the evaluation value S is added, while the familiarity value is lower than the first threshold value or the second When exceeding a threshold value, the structure which deducts evaluation value S is assumed. It is also possible to change the added point value or the deducted value for the evaluation value S step by step for each range having each threshold as a boundary.

  As exemplified above, also in the third embodiment, the conversation is evaluated in consideration of the related information R of the conversation in addition to the feature amounts of the voice V1 and the voice V2 constituting the conversation. Therefore, similarly to the first embodiment, it is possible to appropriately evaluate the conversation as compared with the configuration in which only the feature amount is reflected in the evaluation. Particularly in the third embodiment, since the speaker information Q (for example, mutual relationship and intimacy) between the users U is used as the related information R, an appropriate evaluation based on the actual relationship between the users U is realized. Is done.

<Fourth embodiment>
FIG. 5 is a configuration diagram of the conversation evaluation apparatus 100 of the fourth embodiment. As illustrated in FIG. 5, the storage device 14 of the conversation evaluation apparatus 100 according to the fourth embodiment stores attribute information A for each user U. The attribute information A is information indicating the attributes (features and properties) of the user U. Information that depends on the sound produced by the user U is particularly suitable as the attribute information A. The attribute information A of the fourth embodiment specifies the sound frequency of the user U. The sound generation frequency is an average pitch (average pitch) of the sound generated by the user U.

  As in the second embodiment, the information generation unit 34 of the fourth embodiment searches the storage device 14 for the attribute information A of each of the user U1 and the user U2 specified by the input of identification information and speaker identification. Then, the related information R including the sound generation frequency designated by each attribute information A is generated. That is, the related information R of the fourth embodiment is information of each user U who carries out the conversation to be evaluated, and, similar to the related information R exemplified in the first to third embodiments, the feature acquisition unit This is an example of information of a type different from the feature quantity extracted by 32.

  In the first to third embodiments, the configuration in which the related information R generated by the information generation unit 34 is reflected in the evaluation of the conversation by the conversation evaluation unit 36 is exemplified. However, in the fourth embodiment, the feature acquisition unit 32 The related information R is reflected in the feature quantity extraction. That is, the feature acquisition unit 32 of the fourth embodiment extracts feature amounts under conditions according to the related information R generated by the information generation unit 34.

  Specifically, the feature acquisition unit 32 extracts an acoustic component within a predetermined band including the sounding frequency of the user U1 specified by the related information R from the audio signal X1 generated by the sound collection device 22, and after the extraction The pitch P1 is specified from the acoustic component. That is, the pitch P1 is specified only in the sound range that the user U1 normally generates. Similarly, the feature acquisition unit 32 specifies the pitch P2 from the sound component within a predetermined band including the sounding frequency of the user U2 specified by the related information R in the audio signal X2 generated by the sound collection device 24. It is also possible to use the attribute information A that specifies the user U's pronunciation range as the related information R.

  The conversation evaluation unit 36 evaluates the conversation between the user U1 and the user U2 according to the pitch P1 and the pitch P2 specified by the feature acquisition unit 32 using the related information R as illustrated above. Specifically, the conversation evaluation unit 36 calculates the evaluation value S according to the pitch difference ΔP between the pitch P1 and the pitch P2. The related information R is not added to the evaluation by the conversation evaluation unit 36 in the fourth embodiment. However, as in the examples of the first to third embodiments, the related information R can be added to the evaluation by the conversation evaluation unit 36 in the fourth embodiment.

  As illustrated above, in the fourth embodiment, as in the first embodiment, the conversation between the user U1 and the user U2 is objective according to the feature amounts of the voice V1 and the voice V2 constituting the conversation. It is possible to evaluate. Further, in the fourth embodiment, since feature quantities (pitch P1, pitch P2) are extracted under conditions according to the related information R related to conversation, the comparison is made with a configuration that does not use the related information R for feature quantity extraction. Thus, there is an advantage that the feature amount can be appropriately extracted. For example, in the fourth embodiment, it is possible to extract feature quantities with high accuracy by limiting to a frequency band corresponding to the sound generation frequency specified by the related information R.

  In the above example, the pronunciation frequency of the user U is exemplified as the attribute information A, but the content of the attribute information A is not limited to the above example. For example, it is possible to use attribute information A that specifies the sex of the user U. The feature acquisition unit 32 specifies the pitch P within the frequency band assumed for the sex specified by the related information R. For example, when the sex of the user U1 specified by the related information R is female, the feature acquisition unit 32 extracts the pitch P1 from the sound component of the high frequency range expected for the female from the audio signal X1, When the gender of the user U1 is male, the pitch P1 is extracted from the sound component of the bass range assumed for males in the audio signal X1.

  When the pitch difference between the voice V1 of the user U1 and the voice V2 of the user U2 exceeds 1 octave, one of the pitch P1 and the pitch P2 is brought closer to the other by an integer multiple of 1 octave. Therefore, it is preferable to correct the pitch difference between the two within one octave (hereinafter referred to as “pitch correction”). When the gender is different between the user U1 and the user U2 (that is, when the pitch difference is large), it is estimated that the necessity for pitch correction is high. Considering the above tendency, if the gender specified by the related information R is different between the user U1 and the user U2, the feature acquisition unit 32 executes pitch correction, and the user U1 and the user U2 A configuration in which the feature acquisition unit 32 omits pitch correction when the gender is common is also suitable.

<Fifth Embodiment>
As in the third embodiment, the information generation unit 34 according to the fifth embodiment refers to the speaker information Q stored in the storage device 14 to thereby determine the familiarity between the user U1 and the user U2. The related information R to be specified is generated. As in the fourth embodiment, the feature acquisition unit 32 uses the feature amounts (pitches) of the voice V1 of the user U1 and the voice V2 of the user U2 under conditions according to the related information R generated by the information generation unit 34. P1, P2) is extracted. Specifically, the feature acquisition unit 32 extracts feature amounts at a frequency according to the familiarity specified by the related information R.

  For example, when the intimacy is high, the user U1 and the user U2 are in a good relationship, so the evaluation value S is expected to be a relatively large numerical value. When the familiarity is low, it is difficult to predict the magnitude of the evaluation value S. Therefore, it is assumed that the necessity of frequently evaluating conversations is low when the intimacy is high, and that it is necessary to frequently evaluate conversations when the intimacy is low. In consideration of the above tendency, the feature acquisition unit 32 of the fifth embodiment decreases the frequency of extraction of feature amounts (pitch P1, pitch P2) as the familiarity specified by the related information R is higher. .

  Specifically, the feature acquisition unit 32 extracts a feature amount at a lower frequency when the familiarity exceeds a predetermined threshold than when the familiarity falls below the threshold. For example, when the intimacy is lower than the threshold value, it is once for each set of utterance sections adjacent to each other of the voice V1 and the voice V2 (that is, once for each pair of the utterance by the user U1 and the response by the user U2). Pitch P1 and pitch P2 are extracted (in frequency). On the other hand, if the intimacy exceeds the threshold, the frequency of speech V1 and speech V2 is set to a plurality of sets of speech segments (that is, once for each of a plurality of times of a speech by the user U1 and a response by the user U2). ) Pitch P1 and P2 are extracted. Since the evaluation of the conversation by the conversation evaluation unit 36 is performed every time the feature amount is extracted by the feature acquisition unit 32, the higher the familiarity specified by the related information R, the higher the frequency of evaluation by the conversation evaluation unit 36 (and also the display) The frequency of updating the evaluation value S displayed on the device 16 decreases. A plurality of different threshold values can be used. For example, a configuration is assumed in which the frequency is set for each range having each of a plurality of thresholds as a boundary, and the feature acquisition unit 32 extracts the feature amount at a frequency corresponding to the range to which the familiarity belongs among the plurality of ranges.

  As illustrated above, in the fifth embodiment, as in the first embodiment, the conversation between the user U1 and the user U2 is objective according to the feature amounts of the voice V1 and the voice V2 constituting the conversation. It is possible to evaluate. Further, in the fifth embodiment, feature amounts (pitch P1, pitch P2) are extracted under the conditions according to the related information R. Therefore, as in the fourth embodiment, the related information R is extracted for feature amount extraction. There is an advantage that the feature amount can be appropriately extracted as compared with the configuration that is not used. For example, in the fifth embodiment, since the frequency of feature quantity extraction is controlled according to the related information R, the feature quantity obtained by the feature acquisition unit 32 is compared with a configuration that does not use the related information R for feature quantity extraction. It is possible to reduce the amount of calculation required for extraction and conversation evaluation by the conversation evaluation unit 36.

  In the illustration of the fifth embodiment, the feature quantity extraction condition (specifically, the frequency) is controlled according to the familiarity between the users U, but the content of the related information R to be reflected in the feature quantity extraction condition. Is not limited to the above examples. For example, any related information R exemplified in the first to third embodiments can be applied to control of the feature amount extraction condition. For example, the configuration for controlling the feature quantity extraction condition according to the conversation frequency and the relation period exemplified in the second embodiment (for example, the higher the conversation frequency or the longer the relation period, the more the feature quantity extraction frequency is A configuration to be reduced) is also assumed.

  In addition, the configurations of the first to third embodiments in which the related information R is added to the conversation evaluation by the conversation evaluation unit 36 and the feature amount extraction condition by the feature acquisition unit 32 are controlled according to the related information R. It is also possible to merge the configurations of the fourth embodiment and the fifth embodiment. Different types of related information R can be suitably applied to the conversation evaluation and the feature value extraction condition control by the conversation evaluation unit 36, but the related information R can also be applied in common. For example, in a configuration in which the related information R includes the conversation frequency, the feature amount extraction frequency by the feature acquisition unit 32 is controlled according to the conversation frequency, and the conversation evaluation unit 36 evaluates the conversation as illustrated in the second embodiment. It is also possible to divert conversation frequency.

<Modification>
Each aspect illustrated above can be variously modified. Specific modifications are exemplified below. Two or more modes arbitrarily selected from the following examples can be appropriately combined within a range that does not contradict each other.

(1) In each of the above-described embodiments, the conversation composed of the voice V1 sounded by the user U1 and the voice V2 sounded by the user U2 is evaluated, but the voice evaluated by the conversation evaluation device 100 is the user U1. It is not limited to the utterance sound (ie, real voice). Specifically, one of the voice V1 and the voice V2 can be a synthesized voice generated by a known voice synthesis technique. For example, the same configuration as that of each of the above-described embodiments can be adopted for evaluating a conversation composed of the voice V1 generated by the user U1 and the voice V2 generated by voice synthesis. That is, by analyzing the utterance content by voice recognition of the voice U1 of the user U1, a voice V2 having an appropriate response to the utterance of the user U1 is generated. It is also possible to selectively select a plurality of voices recorded in advance as the voice V2. Also, a configuration for evaluating a conversation composed of the speech V1 generated by speech synthesis and a speech V2 generated by the user U2, and a conversation composed of the speech V1 and the speech V2 generated by speech synthesis are evaluated. Configurations can also be employed.

  As described above, in the configuration in which the speech V1 and the speech V2 are generated by speech synthesis, the sound collection device 22 and the sound collection device 24 are omitted. Further, in a configuration using synthesized speech, the feature acquisition unit 32 may acquire speech synthesis parameters that specify acoustic characteristics (for example, pitch and volume) of speech as feature quantities of speech V1 and speech V2. Is possible. In the above configuration, the process of extracting the feature quantity of the voice V1 by analyzing the voice signal X1 and the process of extracting the feature quantity of the voice V2 by analyzing the voice signal X2 can be omitted. As understood from the above description, the feature acquisition unit 32 is comprehensively expressed as an element for acquiring the feature amounts of the voices (V1, V2) constituting the conversation, and the feature amounts are obtained by analysis processing for feature extraction. In addition to elements extracted from audio signals, elements that acquire feature values by any method other than extraction are also included. That is, the feature value “extraction” is an example of the feature value “acquisition”.

(2) In each embodiment described above, the evaluation value S calculated by the conversation evaluation unit 36 is displayed on the display device 16, but the form of the evaluation result by the conversation evaluation unit 36 is not limited to the evaluation value S. For example, an evaluation comment corresponding to the evaluation value S can be displayed on the display device 16 (whether or not the evaluation value S is displayed is irrelevant). Moreover, the output method of an evaluation result is not limited to display. For example, the evaluation value S and the evaluation comment can be output by voice.

(3) The method of calculating the evaluation value S according to the feature amount (pitch P1, pitch P2) and the related information R is not limited to the examples in the above-described embodiments. For example, the evaluation value S can be calculated by an operation (for example, a weighted sum) in which a numerical value obtained by evaluating conversation according to a feature amount and a numerical value calculated according to related information R are applied. Further, for example, even in a configuration in which the relationship between the feature quantity and the evaluation value S (for example, the type or coefficient of an arithmetic expression that defines the relationship between the two) is changed according to the related information R, both the feature quantity and the related information R are used. A corresponding evaluation value S can be calculated.

(4) The feature quantity extracted by the feature acquisition unit 32 is not limited to the pitch (P1, P2). For example, the feature acquisition unit 32 can extract the volume of each of the voice V1 and the voice V2 as a feature amount. The conversation evaluation unit 36 evaluates the conversation according to the volume difference between the voice V1 and the voice V2, for example. For example, the conversation evaluation unit 36 calculates the evaluation value S so that the evaluation value S increases as the volume difference between the voice V1 and the voice V2 approaches a predetermined value.

  It is also possible for the feature acquisition unit 32 to extract the interval between the speech segment of the voice V1 and the speech segment of the voice V2 (hereinafter referred to as “speech interval”) as a feature amount. When the speech interval at the time of conversation is appropriate, there is a tendency that the voice of the conversation partner is perceived as a good impression with a sense of security. Considering the above tendency, a configuration in which the conversation evaluation unit 36 calculates the evaluation value S so that the evaluation value S becomes a larger value as the utterance interval is closer to the predetermined value is preferable.

(5) The method by which the information generation unit 34 generates the related information R is not limited to the examples of the above-described embodiments. Specifically, the information generation unit 34 can generate the related information R from the result of analyzing the audio signal X1 and the audio signal X2. For example, using the result of the feature acquisition unit 32 specifying the pitch P1 of the voice V1 and the pitch P2 of the voice V2, the information generation unit 34 estimates the gender of each of the user U1 and the user U2. Similarly to the fourth embodiment, it is also possible to generate the related information R specifying the sexes of the user U1 and the user U2.

(6) The conversation evaluation device 100 can be realized by a server device (a single device or a server system including a plurality of devices) that communicates with a terminal device such as a mobile phone or a smartphone. For example, the conversation evaluation device 100 receives the audio signal X1 and the audio signal X2 from the terminal device, and evaluates the conversation between the user U1 and the user U2 in the same manner as the above-described embodiments (for example, the evaluation value) S) is transmitted to the terminal device.

(7) As described above, the conversation evaluation device 100 exemplified in each of the above embodiments can be realized by the cooperation of the control device 12 and a program. For example, the program of the first aspect corresponding to the first to third embodiments is a feature acquisition for acquiring a feature amount of speech constituting a conversation by a computer (for example, one or more processing circuits) such as the control device 12. The unit 32 functions as an information generation unit 34 that generates related information R that is different from the feature amount, and a conversation evaluation unit 36 that evaluates the conversation according to the feature amount and the related information R.

  Further, the program of the second aspect corresponding to the fourth embodiment or the fifth embodiment is a feature in which a computer (for example, one or a plurality of processing circuits) such as the control device 12 acquires a feature amount of speech constituting a conversation. The acquisition unit 32 is a program that functions as an information generation unit 34 that generates related information R that is different from the feature amount of the conversation, and a conversation evaluation unit 36 that evaluates the conversation according to the feature amount. The feature amount is acquired under conditions according to the related information R.

  The program of each aspect illustrated above can be provided in the form stored in the computer-readable recording medium, and can be installed in the computer. The recording medium is, for example, a non-transitory recording medium, and an optical recording medium (optical disk) such as a CD-ROM is a good example, but a known arbitrary one such as a semiconductor recording medium or a magnetic recording medium This type of recording medium can be included. It is also possible to distribute the program to a computer in the form of distribution via a communication network.

(8) A preferred aspect of the present invention can also be specified as an operation method (conversation evaluation method) of the conversation evaluation apparatus 100 exemplified in each of the above embodiments. In the conversation evaluation method according to the first aspect, a computer (a single computer or a system composed of a plurality of computers) acquires a feature amount of speech constituting a conversation, and the related information R is different from the feature amount of the conversation. And the conversation is evaluated according to the feature amount and the related information R. In the conversation evaluation method according to the second aspect, the computer acquires the feature quantity of the speech constituting the conversation, generates related information R different from the feature quantity for the conversation, and according to the feature quantity and the related information R This is a method for evaluating conversation, and in acquiring feature values, the feature values are acquired under conditions according to the related information R.

DESCRIPTION OF SYMBOLS 100 ... Conversation evaluation apparatus, 12 ... Control apparatus, 14 ... Memory | storage device, 16 ... Display apparatus, 18 ... Input device, 22 ... Sound collection apparatus, 24 ... Sound collection apparatus, 32 ... Feature acquisition part, 34 ... Information generation part, 36 ... Conversation evaluation department.

Claims (7)

A feature acquisition unit for acquiring the feature amount of the speech constituting the conversation;
An information generating unit that generates related information different from the feature amount for the conversation;
A conversation evaluation unit comprising: a conversation evaluation unit that evaluates the conversation according to the feature amount and the related information. The conversation evaluation apparatus according to claim 1, wherein the feature acquisition unit acquires the feature amount under a condition corresponding to the related information. A feature acquisition unit for acquiring the feature amount of the speech constituting the conversation;
An information generating unit that generates related information different from the feature amount for the conversation;
A conversation evaluation unit that evaluates the conversation according to the feature amount;
The feature acquisition unit acquires the feature amount under a condition according to the related information. The feature acquisition unit acquires the pitch of each of the first voice and the second voice constituting the conversation as the feature amount,
The conversation evaluation apparatus according to claim 1, wherein the conversation evaluation unit evaluates the conversation according to a pitch difference between the first sound and the second sound. The related information indicates at least one of a temporal situation of the conversation, a history of past conversations between speakers of the conversation, a relationship between speakers of the conversation, and attributes of each speaker of the conversation. It is information. The conversation evaluation apparatus in any one of Claims 1-4. Computer
A feature acquisition unit for acquiring feature amounts of speech constituting a conversation;
An information generation unit that generates related information different from the feature amount for the conversation; and
A program that functions as a conversation evaluation unit that evaluates the conversation according to the feature amount and the related information. Computer
A feature acquisition unit for acquiring feature amounts of speech constituting a conversation;
An information generation unit that generates related information different from the feature amount for the conversation; and
A program that functions as a conversation evaluation unit that evaluates the conversation according to the feature amount;
The feature acquisition unit is a program for acquiring the feature amount under a condition corresponding to the related information.

Images