JP2013207508A - Automatic voice response device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an automatic voice response device, capable of improving use difficulty on the user side at automatic voice response.SOLUTION: An automatic voice response device 10 includes: a splitter 11 for splitting a voice signal from a user 1 into a background sound and a voice; based on the split background sound, a background state recognition unit 12 for recognizing a background state in which the user 1 is placed; based on the split voice, a user state recognition unit 13 for recognizing a user state, the state of an individual user; based on the recognized background state and the user state, a response content generator unit 14 for generating a response content to respond to the user 1; and based on the generated response content, a voice synthesizer unit 15 for synthesizing and outputting voice.

Description

  The present invention relates to an automatic voice response device that automatically responds to a phone call from a user with synthesized voice.

  Although an automatic voice response device has been conventionally used in, for example, a call center or the like, it can respond only within very limited options such as caller ID, incoming number, and tone input. In such a case, since the response is a one-way response that does not take into account the user's condition, there are inconveniences such as placing an extra burden on the user or not reaching the result desired by the user. As this type of device, for example, Patent Document 1 describes a telephone reception system. In this system, the degree of dissatisfaction is measured by personal identification based on a voice pattern, but the state other than the voice pattern is not considered, and there is no feedback to the system.

Japanese Patent No. 4067481

  SUMMARY OF THE INVENTION An object of the present invention is to provide an automatic voice response device capable of improving the difficulty of use on the user side as compared with the prior art in automatic voice response.

In order to achieve the above object, the present invention provides the following automatic voice response apparatus.
(1) a splitter that separates an audio signal from a user into background sound and audio;
A background state recognition unit that recognizes a background state of the user based on the background sound;
A user state recognizing unit for recognizing a user state which is a state of the individual user based on the voice;
A response content generator for generating a response content for responding to the user based on the background state and the user state;
An automatic speech response apparatus, comprising: a speech synthesis unit that synthesizes and outputs speech based on the response content.
(2) The background state recognition unit includes a background state database in which a plurality of background sounds are converted into a frequency distribution and the background states related to the background sounds are stored in correspondence with each other, and the frequency distributions of the plurality of stored background sounds And the background coefficient corresponding to the background sound showing the maximum correlation coefficient is recognized as the background condition where the user is placed. The automatic voice response device according to (1), characterized in that it is characterized in that
(3) The automatic voice response according to (2), wherein the background state recognition unit further recognizes a background state where the user is placed based on the magnitude of the separated background sound. apparatus.
(4) The user state recognizing unit stores each of the user frequency states obtained by converting the speech corresponding to a plurality of user states into a frequency distribution and using the center frequency value and the utterance pitch fluctuation value as scores. A score corresponding to the center frequency and utterance pitch obtained by converting the separated speech into a frequency distribution from the user state database, and a score and utterance pitch corresponding to the determined center frequency Any one of the above (1) to (3), wherein a user state corresponding to a voice having a maximum total score corresponding to is recognized as a user state that is the state of the individual user The automatic voice response device described in 1.
(5) The user state database further stores, as a score, a value of sound volume corresponding to a plurality of user states, and uses the score corresponding to the separated sound level. The user state corresponding to the voice that is obtained from the person state database and has the maximum value of the score corresponding to the obtained center frequency, the score corresponding to the utterance pitch, and the score corresponding to the volume of the voice. The automatic voice response device according to (4), wherein the automatic voice response device is recognized as a user state which is an individual state.
(6) In any one of the above (1) to (5), the response content generation unit can perform at least one of raising and lowering a volume and switching between a male sound and a female sound during an automatic voice response. The automatic voice response device described.
(7) The response content generation unit outputs, as the response content, any one of a standard response text, a simplified response text, and a response instruction by an operator, according to any one of the above (1) to (6) The automatic voice response device described.

According to the first aspect of the present invention, it is possible to provide an automatic voice response device capable of improving the difficulty of use on the user side as compared with the prior art in automatic voice response.
According to the invention which concerns on Claim 2, the background state in which the user is placed can be recognized easily.
According to the invention which concerns on Claim 3, the response according to the background state where the user is placed can be performed.
According to the invention which concerns on Claim 4, the user state which is a user's individual state can be recognized easily.
According to the invention which concerns on Claim 5, the user state which is a user's individual state can be recognized still more easily.
According to the invention which concerns on Claim 6, the response according to the background state in which the user is set | placed, or a user's individual state can be performed.
According to the invention which concerns on Claim 7, it can respond flexibly with respect to the user state which is a user's individual state.

It is a figure for demonstrating one Example of the automatic voice response apparatus which concerns on this invention. (A)-(d) is a figure for demonstrating the structural example of a background state recognition part. (A)-(d) is a figure for demonstrating the structural example of a user state recognition part. (A)-(c) is a figure for demonstrating the structural example of a response content production | generation part. It is a figure which shows an example of the response flow in the example of FIG.

  FIG. 1 is a diagram for explaining an embodiment of an automatic voice response apparatus according to the present invention. In this example, a call center system will be described as an example, but the present invention is not limited to this. As shown in the figure, when a user 1 first makes a call to a call center, the voice signal of the user 1 is sent to the call center via a PSTN (Public Switched Telephone Networks) 2 which is a general subscriber telephone line network. Received by the telephone line interface unit 3. The automatic voice response device 10 uses a predetermined synthesized voice or voice generated by the operator 16 in response to an utterance (voice signal) from the user 1 input from the telephone line interface unit 3, and the telephone line interface unit 3 and Responds to user 1 via PSTN2.

  The automatic voice response apparatus 10 includes a splitter 11 that separates a sound signal from the user 1 into background sound and sound, and a background that recognizes the background state of the user 1 based on the separated background sound. Based on the state recognition unit 12, the user state recognition unit 13 that recognizes the user state that is the individual user state based on the separated voice, and the user based on the recognized background state and user state 1 includes a response content generation unit 14 that generates response content for responding to 1 and a speech synthesis unit 15 that synthesizes and outputs speech based on the generated response content (text).

For example, the automatic voice response apparatus 10 operates as follows. That is, the automatic voice response apparatus 10 recognizes the background state of the user 1 and the individual state of the user based on the voice signal from the user 1. From the background state and the user state, the response content that can be estimated to be the most “suitable” is selected and used as the output of the automatic voice response device. Here, the response contents that can be estimated to be the most “suitable” include, for example, the following.
・ If the background sound is loud, adjust the volume of the synthesized voice to increase.
-If the user seems angry, adjust the response flow so that the operator responds instead of the response by synthesized speech.
-If the user seems to be in a hurry, adjust the response flow to simplify the response by synthesized speech.
The above response content is an example, and the present invention is not limited to this.

On the other hand, the operator 16 can monitor at least one of the user's voice and synthesized voice as necessary. At this time, when the voice (conversation) of the user 1 flows in an unexpected direction, the operator 16 controls the operation of the response content generation unit 14 or the voice synthesis unit 15 to correct the conversation flow or automatically. You can stop the conversation by answering and switch to your own conversation. As will be described later, as the response content from the response content generation unit 14, an instruction for switching the call may be output to the operator 16 instead of the synthesized speech. This is performed from the unit 14 to the operator 16 via the telephone line interface unit 3 and the voice synthesis unit 15.
Hereinafter, the configuration and operation of each part of the automatic voice response apparatus 10 will be described.

  2A to 2D are diagrams for explaining a configuration example of the background state recognition unit. The background state recognition unit 12 includes background state recognition means 121 and a background state database 122 as shown in FIG. The background state database 122 is obtained by converting a plurality of background sounds into a frequency distribution (frequency spectrum) and storing the background states related to the background sounds in association with each other. The background state recognition unit 121 obtains a correlation coefficient between the stored frequency distribution of the plurality of background sounds and the frequency distribution of the background sound separated by the splitter 11, and obtains the background sound indicating the maximum correlation coefficient. The corresponding background state is recognized and output as the background state where the user is placed.

  FIG. 2B shows a conceptual diagram of the background sound source signal 123 and its frequency distribution 124. FIG. 2C shows the relationship between the frequency distribution of a plurality of background sounds stored in the background state database 122 and the corresponding background state. The frequency distributions 125, 126, and 127 in the figure correspond to the vehicle state, rain, and downtown as background states, respectively. The frequency distribution is normalized. FIG. 2D is a diagram showing a correlation coefficient between the background sound frequency distribution 124 separated by the splitter 11 and the background sound frequency distributions 125, 126, and 127 stored in the background state database 122. The maximum correlation coefficient in the figure is 0.95 of the frequency distributions 124 and 125. In this case, “in the vehicle” corresponding to the frequency distribution 125 in FIG. 2C is recognized as the background state where the user 1 is placed.

  The background state recognition unit 12 can recognize the background state of the user from another viewpoint based on the magnitude (level) of the background sound separated by the splitter 11. For example, if the background sound level is greater than or equal to a predetermined threshold, the background state where the user is placed is “noisy”, and if the background sound level is less than the above threshold, the user is placed. Can be recognized as "silence".

  3A to 3D are diagrams for explaining a configuration example of the user state recognition unit. As shown in FIG. 3A, the user state recognition unit 13 includes a user state recognition unit 131 and a user state database 132. The user state database 132 stores the values of the center frequency and the utterance pitch fluctuation values obtained by converting speech corresponding to a plurality of user states into frequency distributions as scores. The user state recognizing means 131 obtains a score corresponding to the center frequency and speech pitch obtained by converting the voice separated by the splitter 11 into a frequency distribution from the user state database 132, and uses the obtained center frequency as the obtained center frequency. The user state corresponding to the voice having the maximum sum of the corresponding score and the score corresponding to the utterance pitch is recognized and output as the user state that is the individual user state.

  The user state database 132 can further store a value of a sound volume (level) corresponding to a plurality of user states as a score. In this case, the user state recognizing unit 131 obtains a score corresponding to the loudness of the sound separated by the splitter 11 from the user state database 132, and corresponds to the score and speech pitch corresponding to the center frequency obtained above. It is possible to recognize and output the user state corresponding to the voice having the maximum value of the score and the score corresponding to the volume of the sound obtained this time as the user state that is the individual user state.

  FIG. 3B shows a conceptual diagram of the audio original signal 133 and its frequency distribution 134. FIG. 3C shows a center frequency (kHz) obtained by converting voice corresponding to a plurality of user states (male, female, emergency, and angry) stored in the user state database 132 into a frequency distribution. , The fluctuation value of the utterance pitch (%), and the utterance level (dB) (speech size), respectively, are scored. For example, as shown in the shaded portion in FIG. 3C, the center frequency obtained by converting the original audio signal 133 into the frequency distribution 134 is 1.8 kHz, the utterance pitch is 0%, and the utterance level is − In the case of 20 dB, the total score of each user state is as shown in FIG. That is, as a user state, “male” has a total score of 30, “female” has a total score of 10, “emergency” has a total score of 0, and “angry” has a total score of 10. In this case, “male” having the maximum total score is recognized as the user state of the individual user. In this example, the center frequency, the utterance pitch, and the utterance level are obtained as follows, for example. The center frequency is the peak frequency in the frequency distribution obtained by converting the voice. The utterance pitch is the transition frequency of the center frequency per fixed time. The utterance level is obtained by comparing with the volume of the time signal that is a constant level.

  4A to 4C are diagrams for explaining a configuration example of the response content generation unit. The response content generation unit 14 includes a response content generation unit 141 and a response direction database 142 as illustrated in FIG. As shown in FIG. 4B, the response direction database 142 includes A1: volume (up), A2: volume (down), B1: speaker switch (male sound), B2: speaker switch ( Female sound), C1: content (operator), C2: content (standard), C3: content (simplified). The background state (noisy, quiet, in the car, rain, downtown) and the user state (male, female, emergency, angry) corresponding to each response direction are represented by scores according to each state. ing. 4B, the background states output from the background state recognition unit 12 are “noisy” and “downtown”, and the user state output from the user state recognition unit 13 is “ It is urgent.

  The response content generation unit 141 generates response content (text) as follows and outputs it to the speech synthesizer. As shown in FIG. 4C, the total score in each response direction is “A1: volume (up)” is 70, “A2: volume (down)” is 0, “B1: speaker switching (male sound) ) "Is 0," B2: Switch speaker (female sound) "is 0," C1: Content (operator) "is 30," C2: Content (standard) "is 0, and" C3: Content (simplified) " 100. In the comparison between A1 and A2, since the total score of A1 shows the maximum value, “A1: Volume (up)” is selected as the response content. In the comparison between B1 and B2, the sum of both scores shows the same value, so the speaker switching is not performed as the response content, the male sound is the same as the previous male sound, and the female sound is the same as the previous female sound. Is done. In the comparison between C1, C2 and C3, the total score of C3 shows the maximum value, so “C3: content (simplified)” is selected as the response content. Note that when the sum of the scores in the response directions A1 and A2 is the same value, the volume can be changed. Moreover, when the sum total of the score of response direction C1-C3 is the same value, a response direction can be made unchanged with the last time.

  As described above, the response content generation unit 14 can perform at least one of raising and lowering the volume of the synthesized voice and switching between male sound and female sound during automatic voice response. Further, the response content generation unit 14 sets “C2: content (standard)” (standard response text), “C3: content (simplified)” (simplified response text), and “C1: content (operator)” as the response content. Any one of (response instruction by the operator) can be output.

  FIG. 5 is a diagram illustrating an example of a response flow in the example of FIG. First, the response content generation means, in step 51, the response content is “Thank you for calling. Can be used as text. In this case, as described above, since “volume (up)” 56 of A1 is selected for the response directions A1 and A2, an instruction to increase the volume of the synthesized speech is included as the response content. In addition, as described above, since the scores in the response directions B1 and B2 are the same, “no speaker switching” 57 is set, and as a response content, in the case of the previous male sound, the synthesized voice is directly used as a male sound, and the previous female In the case of a sound, an instruction for a female sound is included.

  Next, in step 52, “content (simplified)” 58 of C3 is selected for the response direction C1-C3 as described above. ”Is generated as a text for an inquiry about a purchased product. If “content (standard)” of C2 is selected, the response content is “Please give me a request” in step 54. If the inquiry is about a purchased product, 2 is generated as a text, and if “content (operator)” of C1 is selected, an instruction “switch call to operator” in step 55 is given as the response content Is output. Depending on the type of response direction, it may be applied midway. For example, in the previous step 51, the volume can be raised while speaking “Thank you for calling ...”.

  As described above, in the automatic voice response apparatus, the voice from the user is received by the telephone line interface unit and then input to the splitter. The splitter separates the background sound and the conversation content (speech) according to the frequency band and spectrum distribution difference of the incoming signal, and inputs them to the background state recognition unit and the user state recognition unit, respectively. The background state recognition unit estimates the user's situation from the background sound level, frequency distribution, etc., and inputs it to the response content generation unit as the background state. The user state recognizing unit estimates a change in the state of the user from the pitch, frequency distribution, speech level, etc. of the user's voice, and inputs it to the response content generating unit as the user state. The response content generation unit generates response content that seems most appropriate from the background state and the user state, and inputs the response content to the speech synthesis unit. The speech synthesizer synthesizes response speech based on the given response content (text) and responds to the user through the telephone line interface unit.

  Thereby, an automatic voice response device that is easy for the user to use can be obtained. In addition, since the number of cases in which the automatic voice response device cannot handle the operator and the operator needs to respond is reduced, the use cost of the operator can be suppressed. Furthermore, since the correspondence capability of the automatic voice response device is improved, convenience for the user is improved.

1 User 2 PSTN (Public Switched Telephone Network)
DESCRIPTION OF SYMBOLS 3 Telephone line interface part 10 Voice automatic response apparatus 11 Splitter 12 Background state recognition part 13 User state recognition part 14 Response content generation part 15 Voice synthesis part 16 Operator

Claims (7)

A splitter that separates the audio signal from the user into background and audio;
A background state recognition unit that recognizes a background state of the user based on the background sound;
A user state recognizing unit for recognizing a user state which is a state of the individual user based on the voice;
A response content generator for generating a response content for responding to the user based on the background state and the user state;
An automatic speech response apparatus, comprising: a speech synthesis unit that synthesizes and outputs speech based on the response content.   The background state recognition unit includes a background state database that converts a plurality of background sounds into a frequency distribution and stores the background states related to the background sounds in correspondence with each other, and stores the frequency distributions of the plurality of background sounds and the separation And obtaining a correlation coefficient with the frequency distribution of the background sound, and recognizing a background state corresponding to the background sound showing the maximum correlation coefficient as the background state where the user is placed. The automatic voice response device according to claim 1.   The automatic voice response apparatus according to claim 2, wherein the background state recognition unit further recognizes a background state where the user is placed based on the magnitude of the separated background sound.   The user state recognizing unit includes a user state database that stores, as scores, a center frequency value and a speech pitch variation value obtained by converting speech corresponding to a plurality of user states into a frequency distribution. The score corresponding to the center frequency and speech pitch obtained by converting the separated speech into a frequency distribution is obtained from the user state database, and corresponds to the score and speech pitch corresponding to the obtained center frequency. The automatic voice response according to any one of claims 1 to 3, wherein a user state corresponding to a voice having a maximum total score is recognized as a user state which is the individual state of the user. apparatus.   The user state database further stores, as a score, a loudness value corresponding to a plurality of user states, and a score corresponding to the separated sound volume is stored in the user state database. The user status corresponding to the voice showing the maximum value of the score corresponding to the obtained center frequency, the score corresponding to the utterance pitch, and the score corresponding to the volume of the voice is determined as the personal status of the user. The automatic voice response apparatus according to claim 4, wherein the automatic voice response apparatus is recognized as a user state.   The said response content production | generation part can perform at least one of the raising / lowering of a volume at the time of an automatic voice response, and the switch of a male sound and a female sound, The one in any one of Claims 1-5 characterized by the above-mentioned. Automatic voice response device.   The automatic response device according to claim 1, wherein the response content generation unit outputs any one of a standard response text, a simplified response text, and a response instruction by an operator as the response content. .

Images