JP2000194386A - Voice recognizing and responsing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make a voice recognizing and responsing device used by a user more easily and also to make the user perform dialogues which are more efficient and natural with respect to the device. SOLUTION: When voice information are judged to be one of character information by a recognized result judging device 14, the skill of the user is judged by a skill judging device 15 and also the environmental degree of the user is judged by an environmental degree judging device 16. Then, a voice guide selecting device 18 selects the class and the speed of the voice guide by referring to a voice guide dictionary 21, based on the skill judged by the skill judging device 15 and the environmental degree judged by the environmental degree judging device 16 and the controlled result of a dialogue controller 17.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition response apparatus and method using a telephone, and more particularly to a speech recognition response apparatus and method which can be used easily by a user and enable efficient natural conversation.

[0002]

2. Description of the Related Art In recent years, a voice recognition response device for purchasing an article or making a reservation for an air ticket while interacting with a machine using a telephone has become widespread.

Here, a case will be described in which, for example, a flight ticket is reserved using this apparatus. As an example, the departure place is Haneda, the arrival place is Sapporo, and the departure date is December 12, 1998.
When booking a flight ticket at 5:30 pm on the 31st of the month, interact with the device as follows.

There are three methods (1), (2) and (3) below.

In the following description, the utterance as the system prompt (the utterance on the device side) is shown in "", and the utterance of the user corresponding thereto is shown in "".

(1) A question and answer with the device is made in accordance with a voice guide of the fixed device, and each input item is confirmed one by one.

For example, as shown in FIG. 9 (a), in response to the question "Please tell me the departure place", "Haneda" is replied, and "Is the departure place at Haneda Airport?" In response to the confirmation of `` Yes '', in response to the question `` Please say your destination '', reply `` Sapporo '' and in response to the confirmation of `` Is the destination you like at Sapporo Airport? '' ,
Confirm "yes" and respond to the question "Please tell me the date of departure" in the same way.

(2) Basically, the answer is one by one.
The input items such as "year", "month" and "day" and "departure place" and "arrival place" are collectively input and confirmed to some extent.

For example, as shown in FIG. 9 (b), in response to the question "Please tell me the place of departure", "Haneda" is replied, and "Where do you go from Haneda Airport?""Sapporo" is answered in response to the question, and "Yes" is confirmed in response to the two items of "Is it OK to depart from Haneda Airport and go to Sapporo Airport?"

(3) The necessary items are input at a time, and are collectively confirmed later.

For example, as shown in FIG. 9 (c), "Please state the departure place, destination, departure date and time."
Asked, "Hane So Sapporo, 1998 January
Reply "30 minutes afternoon on February 31" and confirm "Yes" in response to "Departure from Haneda Airport to Sapporo Airport, 30 minutes afternoon on December 31, 1998." .

[0012]

However, in the above-described conventional voice recognition response device, the flow of voice guidance and dialogue of the device is fixed irrespective of the user and the situation of the user, and the same voice and the same tone are always present. , With the same audio guide,
Repeat the same exchange in the same order.

Accordingly, the above-mentioned cases (1), (2) and (3) have the following problems.

(1) If a recognition error continues, the same dialogue must be repeated many times. For example, as shown in FIG.
Replying to "Haneda", replying "No" to the confirmation of "Is the departure place at the airport?" It is splashing
"Do you want to start at Hakodate Airport?"
In response to the confirmation, "No" is replied, and the same exchange is repeated, such as "I do not recognize. Please answer" Yes "" No "."

When the same word is repeatedly spoken, the user speaks in a delimited manner like "ha-ne-da",
They tend to pronounce slowly, such as being stretched out like "ha-ne-da", which makes it harder to make speech recognition, and as a result, causes incorrect recognition to be repeated.

(2) In the case of one-by-one answer, the number of utterances of the user increases and the time becomes longer. For example, as shown in Fig. 10 (b), in response to the question "Please tell me the departure place", reply "Haneda" and ask the question "Where do you go from Hata Airport?" On the other hand, it becomes “……”, and the question is asked again, “Would you like to go from Hakata Airport”.

For example, as shown in FIG. 11 (c), in response to the question "Please tell me your departure place", "Haneda" is replied, and "Where do you go from Hata Airport?" To the question "Do you want to depart from Hakata Airport to Sapporo Airport?" And reply "No." Are you sure? "
Return to the first question, "Please tell me where you started."

As described above, even if it is made to input for each word so that even a beginner can understand easily, and it is confirmed with "Yes" or "No" for each input, an inexperienced user can surely input necessary items. However, compared to talking with a human operator, an unnatural conversation with the machine must be performed for a very long time, which imposes a time and psychological burden on the user.

(3) Collective input and collective confirmation make the required time longer. For example, as shown in FIG. 11 (d), in response to the question "Please tell me the departure place, arrival place, departure date and time", "Haneda Sapporo.
5:30 pm on December 31, 1998 "
In response to the confirmation, "Is it okay to leave Haneda Airport for Sapporo Airport at 5:50 pm on October 31, 1999?" Are you sure you want to answer "Yes" to the question "Are you sure you want to arrive at the airport" and "Yes" to the question "Do you want to arrive at the airport in Sapporo?"",Answer" No, "followed by" Please tell us the year of departure. "

As described above, when inputting and confirming in a lump, it is necessary to confirm and correct the erroneously recognized portion at the time of erroneous recognition. However, this exchange is very complicated and requires the patience of the user. become. In addition, it is extremely unlikely that a beginner can hear the above-mentioned audio guide and speak utterly without stagnation or mistake, and often becomes silent without knowing what to say. This gives the user a feeling of anxiety and also causes a longer required time.

Although there is a problem common to the above (1), (2), and (3), a voice guide adapted to an unfamiliar user is redundant for an accustomed user. The fixed voice guide requires the user to listen to and understand the voice at the timing and speed of the utterance on the system side, and imposes a large burden on the user as compared with character information statically displayed on the screen. Also, because it is not possible to skip over text information, if a detailed guide for beginners is presented, the user must listen to the audio guide for the same time regardless of whether it is necessary or not, and the experienced user is extremely patient. Will be forced.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a speech recognition response apparatus and a method which enable a user to easily and efficiently perform a natural conversation.

[0023]

According to a first aspect of the present invention, there is provided a voice recognition / response apparatus for recognizing and responding to an utterance made by a caller by telephone. An utterance state detecting means for detecting, and utterance control means for controlling utterance to the other party based on the utterance state of the other party detected by the utterance state detecting means.

According to a second aspect of the present invention, in the first aspect of the present invention, the utterance state detecting means includes a proficiency detecting means for detecting the proficiency of the utterance of the other party, and an environment of the utterance of the other party. Environmental level detecting means for detecting the level.

According to a third aspect of the present invention, in the first aspect of the present invention, the utterance control means controls an utterance content and an utterance speed.

According to a fourth aspect of the present invention, in the second aspect of the present invention, the proficiency level detecting means determines a time required for the other party to respond to the utterance from the apparatus and a response time of the other party. It is characterized in that it is detected on the basis of:

According to a fifth aspect of the present invention, in the second aspect of the present invention, the environmental level detecting means is configured to determine the strength of the environmental noise of the partner, the strength of the utterance of the partner, and the strength of the utterance of the partner. It is characterized in that the degree of environment is detected based on a difference from the strength of environmental noise of the other party.

According to a sixth aspect of the present invention, there is provided a voice recognition and response device for recognizing and responding to an utterance from a partner by telephone, and a voice recognition means for recognizing the utterance of the other party as character information; Proficiency detecting means for detecting the proficiency of the utterance of the other party, environmental level detecting means for detecting the proficiency of the utterance of the other party, and the apparatus Utterance content changing means for changing the utterance content from the user.

According to a seventh aspect of the present invention, in the sixth aspect of the present invention, the speech recognition means includes a recognition candidate extracting means for extracting a plurality of recognition candidates, and a plurality of recognition candidates extracted by the recognition candidate extracting means. And a rule for extracting one fixed candidate from the candidates, wherein one fixed candidate is extracted from the plurality of recognition candidates using the rule.

According to a further aspect of the present invention, there is provided a voice recognition response method for recognizing and responding to an utterance made by the other party by telephone, wherein the step of detecting the utterance state of the other party includes the steps of: Controlling the utterance to the other party based on the utterance state.

According to a ninth aspect of the present invention, in the invention of the eighth aspect, the utterance state is detected by detecting a proficiency level of the utterance of the other party, and further detecting an environmental level of the utterance of the other party. It is characterized by the following.

According to a tenth aspect of the present invention, in the eighth aspect of the present invention, the utterance control includes controlling utterance content and utterance speed.

According to an eleventh aspect of the present invention, in the ninth aspect of the present invention, the detection of the proficiency level is based on a time until the other party responds to the utterance from the apparatus and a response time of the other party. It is characterized in that it is detected on the basis of:

According to a twelfth aspect of the present invention, in the ninth aspect of the invention, the environmental level is detected by determining the strength of environmental noise of the other party, the strength of speech of the other party, and the strength of speech of the other party. It is characterized in that the degree of environment is detected based on a difference from the strength of environmental noise of the other party.

According to a thirteenth aspect of the present invention, in the voice recognition response method for recognizing and responding to a speech from the other party by telephone, a step of recognizing the speech of the other party as character information, Detecting the proficiency of the utterance on the side, detecting the environmental level of the utterance of the partner, and changing the content of the utterance from the device based on the proficiency and environment of the utterance of the partner. And the following.

According to a fourteenth aspect of the present invention, in the thirteenth aspect, the voice recognition includes the step of extracting a plurality of recognition candidates and the step of extracting one fixed candidate from the plurality of the extracted recognition candidates. And a rule for extracting
One fixed candidate is extracted from the plurality of recognition candidates using the rule.

The invention according to claim 15 is a speech recognition response device which recognizes and responds to a speech from the other party by telephone, wherein the speech content recognition means for recognizing the speech content of the other party, and the speech content recognition means. Utterance content confirmation means for confirming the utterance content of the other party recognized by the means, and utterance content change means for changing the utterance content to the other party based on the confirmation result of the utterance content confirmation means. And

According to a sixteenth aspect of the present invention, in the fifteenth aspect, the utterance content confirmation means includes a recognition target candidate detecting means for detecting a plurality of recognition target candidates, and a related recognition target including the recognition target. And a related recognition target obtaining means for obtaining the recognition target candidate detected by the recognition target candidate detection means when the utterance content of the other party cannot be confirmed by the utterance content recognition means. The utterance content of the other party is specified by matching with the related recognition target including the recognition target acquired by the recognition target acquisition unit.

According to a seventeenth aspect of the present invention, in the invention of the fifteenth aspect, the utterance content confirming means replaces the utterance of the other party with “Nana” when the utterance of “7” is pronounced as “7”. And confirm it.

According to an eighteenth aspect of the present invention, in the invention of the fifteenth aspect, the utterance content confirmation means has utterance storage means for storing the utterance of the other party, and confirms the utterance content of the other party at least a predetermined number of times. If it is not possible, the utterance content to be confirmed includes the utterance of the other party stored above.

According to a nineteenth aspect of the present invention, in a voice recognition responding apparatus that utters a voice including a number in response to a voice uttered by a partner, 1 digit voice prompt from 0 to 9 is stored. Digit voice prompt storage means, 0
A two-digit voice prompt storage means for storing two-digit voice prompts from 0 to 99, wherein even numbers are voiced using only the two-digit voice prompts stored in the two-digit voice prompt storage means. An utterance is generated, and an odd number is generated by using only one one-digit voice prompt stored in the one-digit voice prompt storage means.

According to a twentieth aspect of the present invention, there is provided a voice recognition response apparatus for extracting a speech item of a caller from a plurality of items based on speech from a caller. Registering means, second registering means for dividing and registering a plurality of items registered by the first registering means for specific attributes having relevance, and relevance extraction for extracting the relevance Means, and when the utterance item of the other party cannot be extracted by the first registration means, the relevance extraction means extracts a relevance in a specific attribute including the utterance item, and The utterance item of the other party is extracted from the second registration means based on the relevance of the attribute of the other party.

Also, the invention of claim 21 is a voice recognition response device which recognizes and responds to a speech from the other party by telephone, wherein the volume detection means for detecting the volume of the other party from the speech of the other party; Response volume control means for controlling the volume of the response based on the volume of the other party detected by the volume detection means.

According to a twenty-second aspect of the present invention, in the invention of the twenty-first aspect, the response volume control means increases the response volume as the volume of the partner increases, and decreases the response volume as the volume of the partner decreases. It is characterized by doing.

Also, the invention according to claim 23 is a voice recognition response method for recognizing and responding to a speech from the other party by telephone, wherein the step of recognizing the speech content of the other party includes:
The method further comprises the steps of: confirming the uttered content of the recognized other party; and changing the uttered content to the other party based on the confirmation result.

According to a twenty-fourth aspect of the present invention, in the twenty-third aspect of the present invention, the confirmation of the utterance content includes a step of detecting a plurality of recognition target candidates and a step of acquiring a related recognition target including the recognition target. And, if the utterance content of the other party could not be confirmed, the plurality of recognition target candidates detected in the step of detecting the recognition target candidate and the above-mentioned obtained in the step of obtaining the related recognition target The utterance content of the other party is specified by matching with the related recognition target including the recognition target.

According to a twenty-fifth aspect of the present invention, in the twenty-third aspect of the present invention, the step of confirming the utterance content is performed in a case where the utterance of the other party has a pronunciation of “Shi” indicating “7”. It is characterized in that it is replaced with and confirmed.

According to a twenty-sixth aspect of the present invention, in the invention of the twenty-third aspect, the step of confirming the utterance content stores the utterance of the other party, and if the utterance content of the other party cannot be confirmed more than a predetermined number of times, The utterance content to be confirmed includes the utterance of the other party stored above.

According to a twenty-seventh aspect of the present invention, in the voice recognition response method for uttering a voice including a number in response to a voice uttered by the other party, a one-digit voice prompt from 0 to 9 is stored. Digit voice prompt storage means, 0
A two-digit voice prompt storage means for storing two-digit voice prompts from 0 to 99, wherein even numbers are voiced using only the two-digit voice prompts stored in the two-digit voice prompt storage means. An utterance is generated, and an odd number is generated by using only one one-digit voice prompt stored in the one-digit voice prompt storage means.

According to a twenty-eighth aspect of the present invention, in the voice recognition response method for extracting an utterance item of the other party from a plurality of items based on an utterance from the other party of the telephone, the first of the plurality of items is registered. Registration means, second registration means for dividing and registering a plurality of items registered in the first registration step for each of specific attributes having relevance, and relevance extraction for extracting the relevance And the first step.
If the utterance item of the other party cannot be extracted by the registration means, the relevance extraction step extracts the relevance at a specific attribute including the utterance item, and based on the relevance at this specific attribute, The utterance item of the other party is extracted from the second registration means.

According to a twenty-ninth aspect of the present invention, in the voice recognition response method for recognizing and responding to a speech from the other party by telephone, the volume of the other party is detected from the speech of the other party,
The volume of the response is controlled based on the detected volume of the other party.

According to a thirtieth aspect of the present invention, in the invention of the twenty-ninth aspect, the response volume is controlled such that the response volume increases as the volume of the other party increases, and the response volume increases as the volume of the other party decreases. It is characterized in that it is made smaller.

According to a thirty-first aspect of the present invention, in the voice recognition response device which recognizes a speech from the other party by telephone and responds to the confirmation, the speech of the other party is divided for each confirmation item and a confirmation prompt is issued. A confirmation prompt creating means to be created, an utterance time setting means for setting an utterance time for each confirmation item in the above confirmation prompt, and an input from the other party during the utterance time set by the utterance time setting means Utterance content detecting means for detecting the input utterance content of the other party, and input time detecting means for detecting the input time of the input utterance of the other party. A response of confirmation is made based on the input time of the uttered speech of the other party and the utterance content of the other party detected by the utterance content detecting means.

According to a thirty-second aspect of the present invention, in the thirty-first aspect, the input time of the utterance of the other party detected by the input time detecting means and the utterance of the other party detected by the utterance content detecting means It has a confirmation prompt change means for changing the contents of the confirmation prompt based on the contents.

The invention according to claim 33 is a voice recognition response method for recognizing a speech by recognizing a speech from the other party by telephone, wherein the speech of the other party is divided for each confirmation item and a confirmation prompt is issued. Creating, setting an utterance time for each confirmation item in the confirmation prompt, and, if there is an input from the other party during the utterance time set in the utterance time setting step, the input is made. Detecting the content of the utterance of the other party, and detecting the input time point of the input utterance of the other party,
And responding to the confirmation based on the input time point of the utterance of the other party detected in the detecting step of the input time point and the utterance content of the other party detected in the step of detecting the utterance content. And

According to a thirty-fourth aspect of the present invention, in the thirty-third aspect, the input time point of the utterance of the other party detected in the step of detecting the input time point and the other party detected in the step of detecting the utterance content The content of the confirmation prompt is changed based on the content of the utterance on the side.

[0057]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a first embodiment of a voice recognition response device according to the present invention will be described in detail with reference to the drawings.

In the following description, a case will be described in which the present embodiment is applied to an airline ticket reservation system by telephone. In this system, the name of the airport at the point of departure, the name of the airport at the destination, the date and time of departure are obtained by voice recognition over the telephone, and the corresponding airline ticket is reserved. In an actual system, various functions such as the number of sheets, class information, a payment method, and security are necessary. However, since these functions are not directly related to the purpose of the present embodiment, they will not be described below.

FIG. 1 is a schematic block diagram showing the overall configuration of the voice recognition and response device according to the present embodiment.

In FIG. 1, the voice recognition response device 10
Voice input device 11, voice waveform feature extraction device 12, voice recognition device 13, recognition result determination device 14, proficiency level determination device 1
5, environment degree determination device 16, dialogue control device 17, voice guide selection device 18, waveform feature analysis device 19, voice output device 20, voice guide dictionary 21, recognition dictionary 22, waveform feature database (waveform feature DB) 23, frame A semantic dictionary 24, a recognition result database (recognition result DB) 25, and a dialog history database (dialog history DB) 26;

Here, the voice input device 11, to which the voice of the other party is input, is constituted by a microphone of a telephone receiver or the like, and converts the input voice information into an electrical voice waveform signal.

The speech waveform feature extraction device 12 analyzes the speech waveform from the speech waveform signal obtained by the speech input device 11 and, based on the frequency and the strength of the signal and the variation, changes the number of words of the utterance, the utterance speed, and the utterance speed. Then, the utterance strength, the variation of the utterance strength, and the delay from when the utterance is possible to when the utterance is actually performed are obtained.

The voice recognition device 13 obtains the voice utterance of the other party from the voice waveform signal obtained by the voice input device 11 using character information. The recognition result is not one, but is obtained as some recognition result candidates in a form to which a certainty factor (N-Best) indicating the certainty of the recognition is added. The recognition result is stored in a recognition result database (recognition result DB) 25, and the recognition result is determined by the recognition result determination device 14. Since various methods are already known for the voice recognition processing itself, detailed description thereof will not be given.

The recognition result judging device 14 includes the speech recognition device 1
3 from the recognition result candidates obtained in step 3, the recognition dictionary 22
Then, the recognition result is determined using the frame / semantic dictionary 24. Incidentally, the processing method of this determination will be described later in detail with reference to FIGS.

The proficiency judging device 15 judges a user's proficiency in the present system, and judges the proficiency in three stages of low, medium and high, and, depending on the proficiency, the flow of the dialogue, the contents of the system utterance, and the utterance speed. To adjust. The method for determining the proficiency will be described later in detail with reference to FIG.

The environmental level determination device 16 evaluates environmental variables, such as the voice quality, loudness of the speaker, and the surrounding environment, which affect speech recognition in three stages of excellent, good, and acceptable. And adjust the content of system utterances. The method of determining the degree of environment will be described later in detail with reference to FIG.

The dialogue control device 17 includes the proficiency level determination device 15
It controls the flow of the dialogue with the user based on the determination result of the environment degree determination device 16.

The voice guide selection device 18 determines the selection of the voice guide and the speed of the voice guide based on the determination results of the proficiency level determination device 15 and the environmental level determination device 16. When the proficiency is high, a short and short voice guide is uttered quickly, and when the proficiency is low, a detailed voice guide is uttered slowly. When the user is silent, the user is prompted to re-enter the input, and when the voice input level is low, the user is guided to increase the voice. The method of selecting the voice guide will be described later in detail with reference to FIG.

The waveform characteristic analysis device 19 evaluates the characteristics of the audio waveform based on the extraction result extracted by the audio waveform characteristic extraction device 12, and makes the waveform usable by the proficiency level determination device 15.

The audio output device 20 converts an electric signal into an audio signal, and uses a speaker of a telephone receiver or the like.

The voice guide dictionary 21 holds voice information of voice guides spoken by the device.

The recognition dictionary 22 holds a recognition dictionary used for speech recognition.

Waveform feature database (waveform feature DB) 2
Reference numeral 3 holds the characteristics of the audio waveform extracted by the audio waveform characteristic extraction device 12 and the characteristics of the audio waveform analyzed by the waveform characteristic analysis device 19.

The frame / semantic dictionary 24 holds relations between speech recognition results and semantic constraint information. For example, in the airline ticket reservation frame, there is an airport slot restriction (semantic information) that the same airport name cannot be entered in the departure place slot and the destination slot. Also, 1 to 12 for the month slot
There is a date slot constraint (semantic information) that contains an integer up to. The structure of the frame / semantic dictionary 24 and the method of determining speech recognition using the frame / meaning dictionary 24 will be described later in detail with reference to FIGS.

Recognition result database (recognition result DB) 2
Reference numeral 5 denotes a plurality of recognition result candidates recognized by the voice recognition device 13 and their certainty data, and the recognition result data determined by the recognition result determination device 14 and the determined recognition result candidates by the voice recognition device 13. Holds recognition result data.

Dialogue history database (dialogue history DB) 2
6 is a transition of the proficiency level determined by the proficiency level determination device 15,
The transition data of the environment degree determined by the environment degree determination device 16 and the transition data of the dialogue controlled by the dialogue control device 17 are held.

FIG. 2 is an explanatory diagram of frames and semantic information stored in the frame / semantic dictionary 24. FIG. 2 shows an airline ticket reservation frame 30 for airline ticket reservation and semantic information 31 for airline ticket reservation.

Here, the air ticket reservation frame 30 includes a plurality of slots (items) required for air ticket reservation, ie,
"Departure place", "Arrival place", "Date: Month", "Date:
It has five slots, "day" and "time". Each slot has "Haneda", "Sapporo", "December",
“31st” and “5:50 pm” are described.

The semantic information 31 is composed of constraints on the description contents described in each slot of the airline ticket reservation frame 30.

For example, the "inter-airport slot restriction" 31a has a restriction that (1) the same airport name is not entered and (2) a combination without a flight is not specified.

The "date slot restriction" 31b has the following restrictions: (1) month is an integer from 1 to 12; (2) day is an integer from 1 to 31; (3) a past date is not specified.

As the "time slot constraint" 31c, there is a constraint that (1) hour is an integer from 1 to 12 or 24, and (2) minutes is an integer from 0 to 59.

Therefore, by using the slot constraint included in the semantic information 31 in addition to the recognition result of the speech recognition device 13, the recognition accuracy of the recognition result determination device 14 can be improved.

FIG. 3 illustrates a processing procedure in the case where one recognition result is determined from a plurality of recognition result candidates by the recognition result determination device 14 by using the slot constraint as described above.

FIG. 3 shows a processing procedure when the month of the airline ticket reservation frame 30 is determined, that is, when a “date: month” slot is described.

By the way, when describing the “date: month” slot, of the “date slot constraint” 31b, (1) month is an integer from 1 to 12 and (3) past date is not specified. .

Therefore, if the current date is September 15,
As the user utterance (actual user utterance) 35, "12
When the voice recognition device 13 recognizes the three candidates of February, April, and December with the certainty factor in this order as the recognition process 36, the recognition determination process 37 of the recognition result determination device 14 First, 14 months are deleted with the constraint that the month (1) is an integer from 1 to 12, and then February is deleted with the constraint that the past date is not specified in (3).

Accordingly, the December having the third degree of certainty in the voice recognition of the voice recognition device 13 is passed to the application (application main body) 38 as a confirmation result.

FIG. 4 is a diagram for explaining a proficiency level determination processing method determined by the proficiency level determination device 15.

The skill level is determined based on the utterance recognition result of the user determined by the recognition result determination device 14 and the analysis result of the waveform feature analysis device 19 that analyzes the extraction result of the voice waveform feature extraction device 12.

That is, when the recognition result determination device 14 determines the utterance recognition result of the user, (1) the voice output device 2
Time A from the start of utterance of the voice guide from 0 until the user speaks, (2) User's utterance time T, (3)
Based on the number of uttered sounds (number of uttered words) N of the user obtained from the analysis result of the waveform feature analysis device 19, the times of A and T are short, and the less the number of uttered sounds (number of uttered words) N is, the higher the proficiency level is. , A and T are longer and the number of uttered sounds (number of uttered words) N is larger, the proficiency is determined to be lower.

[0092] This means that a person who is more proficient in the device can quickly determine what to input, and speak without waiting for the end of the voice guide (the time A is short), and the contents of the speech are minimal. This is because there is a tendency to say nothing (the time B is short and the number of uttered sounds (number of uttered words) N is small).

Therefore, A, T, and N are defined as described above, and the number of uttered sounds (the number of uttered words) of the standard user expected by the apparatus is as follows.
Is n, the proficiency level P is obtained by the following equation.

P = A− (N−n + T)

However, A takes the following crisp value. 3: When the user speaks in the first half of the voice guide 2: When the user speaks immediately after the voice guide ends (about 0.2 seconds after the end) from the latter half of the voice guide 1: Whether the user speaks after the voice guide ends , If not spoken

The above is the proficiency level determination method determined by the proficiency level determination device 15.

FIG. 5 is a diagram for explaining a method of determining the environmental level determined by the environmental level determining device 16.

The determination of the environmental level is made based on the user's utterance recognition result determined by the recognition result determination device 14 and the analysis result of the waveform feature analysis device 19 that analyzes the extraction result of the voice waveform feature extraction device 12.

In FIG. 5, indicates the intensity of environmental noise which is input when neither the device side nor the user side is uttering, indicates the user's utterance intensity, and indicates the difference between the user's utterance intensity and the environmental noise intensity. .

Here, the environment degree is small and large, that is, the environment is better as the environmental noise is small and the user's utterance strength is large, and conversely, the environment level is large and small, that is, the environment noise is large and the user's utterance strength is large. It is determined that the smaller the value, the worse the environment. If the value is large even if the value is large, that is, if the value is relatively large, the environment is determined to be good.

Note that the speech signal and noise are separated by FFT.
(High-speed discrete Fourier transform), and the amplitude information in the frequency domain converted by the FFT at an arbitrary time point i is G “i” and the noise amplitude is g “i”.
Each energy is given by the following equation.

Energy E1 in frequency domain including noise
Is: E1 = ΣG “i”^Two  The energy E2 in the frequency domain of the noise is: E2 = Σg “i”^Two  The energy in the voice frequency domain is: E3 = ΣG “i”^Two−Σg “i”^Two

Therefore, the environmental degree S is given by the following equation. S = E3 / E1 × α (where α is a constant)

FIG. 6 is a diagram for explaining a method of selecting a voice guide selected by the voice guide selecting device 18. As shown in FIG.

The selection of the audio guide is performed by the proficiency judgment device 15.
The waveform feature database (waveform feature DB) 23, based on the user's proficiency determined in the above, the environment degree determined by the environment degree determination device 16, and the interaction history of the user and the device controlled by the dialog control device 17. This is performed by referring to data stored in a recognition result database (recognition result DB) 25 and a dialog history database (dialog history database) 26.

That is, the voice guide selecting device 18 is composed of a voice guide rule 40 and a voice guide database (voice guide DB) 41, and the content of the voice guide to be uttered according to the proficiency level, the environmental level and the number of re-inputs. The utterance speed has been changed. For example, if the proficiency level is high, a short guide "Please go to the departure place" is uttered at a high speed. At the time of re-input, the degree of environment is checked. If the environment is bad, a voice guide of the bad environment is uttered, and then a voice guide of the input request is uttered. Also, every time re-input is repeated, the voice guide is changed to a detailed one.

Accordingly, for example, in the case of a voice guide at the departure place, the following rules are prepared as the voice guide rules 40. (1) IF skill level = high THEN voice guide = departure place "0", utterance speed = fast (2) IF skill level = medium THEN voice guide = departure place "1", utterance rate = medium (3) IF skill level = Low THEN Audio guide = Departure point "2", Speech rate = Slow (4) IF 0 <Number of loops <3 THEN Audio guide = Departure point "Loop count" Speech rate = Speed (5) IF Environment = OK THEN Voice Guide = bad environment "0" + voice guide (6) Number of loops + = 1

[0108] The voice guide database (voice guide DB) 41 stores the following voice guide data regarding the departure point request and the bad environment.

As the departure point request “3”, “please specify the departure point”, “please state the airport name of departure point”, “please state the airport name of departure point like“ Haneda Airport ”” "The surroundings or the line are not good. Please talk with the handset a little closer to your mouth."

The contents of the voice guide selecting device 18 have been described above.

Next, the processing procedure of this embodiment will be described with reference to the flowchart of FIG. In the present embodiment, based on the proficiency level determined by the proficiency level determination device 15 and the environmental level determined by the environmental level determination device 16,
Of the three types of audio guidance.

Here, the three types of voice guides are the three types of voice guides (a), (b), and
(C), and as shown in FIG.
The user inputs each item and confirms each item. Also, as shown in FIG. 9B, when two items are input, 2
Items are checked together. Further, as shown in FIG. 9C, all the items are input at once, and thereafter all the items are confirmed.

That is, in FIG. 7, when the program is started, it is checked whether the proficiency level is “high”, “medium” or “low” (step 100).

If the proficiency level is "high", the process proceeds to step 102, where it is checked whether the environmental level is "high", "medium" or "low". Here, the process proceeds to the case where the environmental degree is “high”, and all of the departure place, the destination, the date, and the time are input (step 104), and then the confirmation is performed (step 106). And in this confirmation processing, 1
If there is an error in the item (NO in step 106), the input process in step 104 is performed again. When all the items have been confirmed (YES in step 106), the overall confirmation is performed (step 134), and the process ends.

On the other hand, if it is determined in step 102 that the environmental level is "medium", the processing is performed, and the routine proceeds to step 126. When the environmental level is determined to be “low”, the process is performed, and the process proceeds to step 108.

Next, when the proficiency level is determined to be "low" in step 100, the process of
Proceed to. In step 108, the departure place is first entered,
Subsequently, the confirmation is performed (step 110).

If the departure place cannot be confirmed (NO in step 110), the process in step 108 is performed again. If the departure place can be confirmed (YES in step 110), the process proceeds to step 112. move on.

At step 112, the destination is input, and the confirmation is subsequently performed (step 114).

If the destination cannot be confirmed (NO in step 114), the process in step 112 is performed again. If the destination can be confirmed (YES in step 114), the process proceeds to step 116. move on.

At step 116, the date is subsequently input, and the confirmation is subsequently performed (step 118).

If the date cannot be confirmed (NO in step 118), the process in step 116 is performed again. If the date can be confirmed (step 118).
YES), and proceed to step 120.

At step 120, a time is subsequently input, and then confirmation is made (step 122).

If the time cannot be confirmed (NO in step 122), the process in step 120 is performed again. If the time can be confirmed (step 122).
YES), the process proceeds to step 134, the overall confirmation is performed, and the process ends.

On the other hand, if the proficiency level is determined to be "medium" in step 100, the process proceeds to step 124, where the environmental level is checked.

Here, the process proceeds to the case where the environmental level is "low", and the processes from step 108 onward are performed.

If the environmental level is other than "low", that is, "high" or "medium", the routine proceeds to step 126,
Is performed.

At step 126, a departure place and a destination are first inputted, and then confirmation is made (step 12).
8).

If the starting point and the destination cannot be confirmed (NO in step 128), step 126 is performed again.
Is performed, but if the departure place and the destination are confirmed (YES in step 128), the process proceeds to step 130.

In step 130, the date and time are subsequently input, and the confirmation is subsequently performed (step 13).
2).

If the date and time cannot be confirmed (NO in step 132), the process in step 130 is performed again. If the date and time can be confirmed (YES in step 132), the process proceeds to step 130. Proceeding to 134, the overall confirmation is performed, and this processing ends.

The above is the processing procedure of the present embodiment.

Next, the operation between the components shown in FIG. 1 will be described with reference to FIG.

In FIG. 8, when a voice is input to the voice input device 11, the input voice signal is transmitted to the voice waveform feature extraction device 12 and the voice recognition device 13.

The speech signal sent to the speech waveform feature extraction device 12 is used by the speech waveform feature extraction device 12 for the number of words of speech, speech speed, change in speech speed, speech intensity, speech intensity change, and when speech is possible. , And a delay time until an actual utterance is extracted, and these data are further processed by the waveform characteristic analyzer 19 into a form usable by the proficiency determining device 15. The data processed by the waveform feature analyzer 19 is stored in a waveform feature database (waveform feature DB) 23.

Further, the speech signal sent to the speech recognition device 13 is converted into character information while referring to the recognition dictionary 22, and a plurality of recognition result candidates are obtained.

The recognition result determination device 14 determines one recognition result from the plurality of recognition result candidates obtained by the speech recognition device 13 while referring to the frame / semantic dictionary 24.

The data obtained by the voice recognition device 13 and the recognition result determination device 14 are stored in a recognition result database (recognition result DB) 25.

When the voice information is determined to be one character information by the recognition result determining device 14, the proficiency determining device 15 determines the user's proficiency level and the environmental level determining device 16 determines the user's environmental level. You.

The dialogue control device 17 controls the flow of the dialogue as shown in FIG. 7 based on the proficiency level determined by the proficiency level determination device 15 and the environment level determined by the environment level determination device 16.

The voice guide selecting device 18 refers to the voice guide dictionary 21 based on the proficiency determined by the proficiency determining device 15, the environmental level determined by the environmental level determining device 16, and the control result of the dialog control device 17. To select the type of voice guide and the speed of the voice guide. At this time,
The dialog history between the device and the user is stored in a dialog history database (dialog history DB) 16, and the dialog history stored in the dialog history database (dialog history DB) 16 is used by the voice guide selecting device 18.

The voice guide selected by the voice guide selecting device 18 is output from the voice output device 20.

As described above, in the first embodiment, the skill level and environment of the user are determined, and the voice guide, the dialogue, and the flow of the dialogue are changed. An appropriate response can be made in accordance with the state of the line, a natural conversation without burdening the user can be realized, and the talk time can be shortened.

Therefore, it is possible to provide an interactive device that does not give stress to the user. Further, it is possible to effectively use the telephone line.

By the way, in the conventional voice recognition response device,
There were also the following problems.

(1) It has been considered that a conventional speech recognition response device can realize a smooth conversation if the recognition accuracy is improved. However, the content of unspecified number of people
It is very difficult to recognize 00% correctly. In addition, even if the user's utterance that was once erroneously recognized is repeatedly input, it is not common to understand why the utterance is not recognized, so that the user utters the same utterance. When such an erroneous recognition occurs, the same question is repeated until a correct answer is given, and eventually the operator is connected to tell what he / she wants to say. Such a process does not meet the user's need to convey what he / she wants to convey in as short a conversation time as possible.

(2) Although the amount of utterance of the user greatly affects the recognition accuracy, no countermeasure has been taken conventionally. In addition, if the processing is to calculate a characteristic amount after performing filter processing such as gain adjustment and modulation, the calculated characteristic amount may be slightly shifted, which may adversely affect recognition accuracy.

(3) In the case where a repeat prompt (question) for confirming the recognition result uttered from the device includes a plurality of digits, conventionally, a single-digit number prompt (0 to 9) recorded in advance is used. , The number of digits is uttered to create the prompt, which makes the connection of time unnatural and difficult to hear.

Thus, as a second embodiment, a speech recognition responder adopting the following configuration will be described below. (1) The user does not repeat the same system prompt during the conversation as much as possible to reduce the discomfort generated by the user. (2) Relevant questions are narrowed down without repetition of questions having the same content to narrow down recognition results, thereby increasing recognition accuracy and reducing conversation time. (3) Based on how the user responds, the apparatus works on the user to create a situation that is easy to recognize, and increases the recognition accuracy to reduce the conversation time. (4) It is possible to create a low-cost and easy-to-hear system prompt by devising an unnatural number repeat prompt.

In the following description, a case will be described in which the present embodiment is applied to a voice recognition response device that performs an aircraft departure and arrival guidance service.

The aircraft departure / arrival guidance service has two types of guidance services. One is for the user to enter the flight number of the aircraft to speak the flight arrival / departure information, and the other is for the flight arrival / departure information. By inputting the destination and the destination, information on all the flights operated on that route is uttered.

FIG. 12 is a block diagram showing the overall processing flow of this embodiment.

In the following description, the overall processing flow will be described for each state. The state means a set of a system prompt (utterance from the device side) and a user's utterance (answer) in response to the prompt.

In the following description, the utterance as the system prompt is shown in "", and the utterance of the user corresponding thereto is shown in "".

(201) Service Selection State As described above, in this embodiment, two types of services are provided. In this state, the user is asked which service to use and a response is obtained. .

At the system prompt, the user utters "Which service is used, flight number or route?", And the user returns a response of "flight name" or "route". Then, by this selection, the service is branched. If it can be recognized, the state transitions to the selected service confirmation state 202. If it cannot be recognized, the state transitions to the YesNo state 203.

(202) Selected Service Confirmation State This is the part for returning the selected service to the user at the repeat prompt and obtaining the correctness. The system prompt is "Are you sure with" flight service "?" Or "Are you sure with" route service "?", And the user's response to it is either "yes" or "no."

If the user's response is "yes", the flow proceeds to the recognition result determination unit 204. If the response is "no" or recognition is not possible, the flow proceeds to the YesNo state 203.

(203) YesNo state When the service cannot be recognized in the service selection state 201,
This is a state to which a transition is made when it is found that misrecognition has been made in the selected service confirmation state 202, or when "yes" or "no" cannot be recognized. Here, the system prompt is a state in which the user utters either "is the flight number?" Or "is the route?" And the user can answer "yes" or "no".

That is, if the system prompt is "Flight name?", The user utters "Yes" to know that he / she wants the "flight name information service". Service ".

If the system prompt is "Route line?", If the user speaks "Yes", it is understood that the user is requesting "Route information service". "Is required.

(204) Recognition Result Determining Unit Here, the service used by the user is determined to be either “flight name” or “route” based on the recognition result.

(205) Previous State User Speech Amount Check Unit Here, since the user has made several utterances in the previous state, the magnitude of the utterance amount at that time is detected, and the system prompt volume adjustment unit is detected. Pass it to 206.

(206) System Prompt Volume Adjusting Unit The volume of the system prompt is adjusted based on the user's utterance detected by the pre-state user utterance checking unit 205.

Although the processing procedure in this case will be described later in detail, the volume adjustment is such that the system prompt volume increases as the user's utterance increases, and the system prompt volume decreases as the user's utterance decreases.

(207) Service branching unit Here, a branch is made to the flight number information service and the route information service.

(208) Flight Number Recognition State This is a state to which a transition is made when using the flight number information service. The system prompt here is "Please say the flight number you want to look up", and the user says the flight name you want to look up in either with or without digits.

In this case, if there is a digit, the expression is "Nanahyakusanjugo" bottle, and if there is no digit, the expression is "" Nanasango "bottle".

In the flight number recognition state 208,
"Unrecognizable" means that this state is repeated again when recognition is not possible. The same applies to the following.

(209) Flight number confirmation state At the repetition prompt, the correctness of the flight number recognized in the flight number recognition state 208 is confirmed.

As a result of the confirmation, if the user's response is “yes”, the processing transits to the recognition result deciding section 211, but “no”
In this case, the process proceeds to the digit pronunciation method prompt generation unit 210 or the flight number recognition state 208. The transition to the number pronunciation method prompt generation unit 210 occurs when the number of the recognition result includes either 1 or 7.

(210) Numeric pronunciation method prompt generation unit In the flight number confirmation state 209, the user answers "No",
If 1 or 7 is included in the recognized number, a system prompt of "Please say" Nana "to pronounce the number 7" is uttered.

(211) Recognition result deciding section This section decides the flight number for which the user wants to know the operation status.

(212) Operation status guidance section This section informs the user of the operation status. When the flight name information service is used, the status of arrival and departure of the flight is spoken.

The above is the processing flow when the flight number information service is selected.

(213) Departure Airport Name Recognition State When the route information service is selected, the service branching unit 20
This is the state that transitions from 7. The system prompt here is "Please tell me the departure airport name of the route you want to look up.", And the user will say the departure airport name.

In this departure airport name recognition state,
A plurality of airport names having similar pronunciations are recognized.

(214) Departure Airport Name Confirmation State At the repetition prompt, the airport name with the highest certainty among the plurality of airport names recognized in the departure airport name recognition state 213 is repeated to confirm the correctness of the departure airport name. If the user's answer is "yes", the process proceeds to the recognition result determination unit 217,
If “No”, the state transits to the prefecture name recognition state 215.

(215) Prefecture Name Recognition State When "No" is answered in the departure airport name confirmation state 214, the name of the prefecture where the airport is located is heard. For example, despite the user saying "Itami", "Kitami"
, The system prompt says, "Please state the name of the state or province where the departure airport is located."
The user answers "Osaka".

(216) Airport Name Confirmation State The airport name is confirmed using the recognition result in the prefecture name recognition state 215 and a plurality of recognition candidate names recognized in the departure airport name recognition state 213. .

For example, if the prefecture name recognized in the prefecture name recognition state 215 is “Osaka” and “itami” exists in the recognition candidates recognized in the departure airport name recognition state 213, “Osaka” Is it OK at Itami Airport? " On the other hand, the user can answer “Yes” or “No”
To return the answer. If “yes”, the process transits to the recognition result determination unit 217, and if “no”, the process returns to the prefectural name recognition state 215.

(217) Recognition result determination section This is a section for determining the departure airport name of the route on which the user wants to know the operation status.

(218) Arrival Airport Name Recognition State This is a state to which the state changes when the departure airport name is determined by the recognition result determination unit 217. The system prompt here is "Please state your arrival airport name."
The user will say the arrival airport name.

In this arrival airport name recognition state,
Multiple airport names are recognized.

(219) Arrival Airport Name Confirmation State At the repetition prompt, among the airport names recognized in the arrival airport name recognition state 218, the airport name with the highest certainty is repeated to confirm the correctness of the arrival airport name. . If the user's answer is "yes", the process proceeds to the recognition result determination unit 222,
If “No”, the state transits to the prefecture name recognition state 220.

(220) Prefecture Name Recognition State When "No" is answered in the arrival airport name confirmation state 219, the name of the prefecture where the airport is located is heard. For example, despite the user saying "Kitami", "Itami"
Is recognized, a system prompt is spoken saying "Please state the name of the state or province where the arrival airport is located."
The user will answer "How are you?"

(221) Airport Name Confirmation State The airport name is confirmed using the recognition result in the prefecture name recognition state 220 and a plurality of recognition candidate names recognized in the arrival airport name recognition state 218. .

For example, if the prefecture name recognized in the prefecture name recognition state 220 is “Hokkaido” and “Kitami” exists in the recognition candidates recognized in the arrival airport name recognition state 218, “ Is it OK at Kitami Airport in Hokkaido? " In response, the user answers “yes” or “no”. Then, in the case of “Yes”, the process transits to the recognition result determination unit 222, and in the case of “No”, the process returns to the prefectural name recognition state 220.

(222) Recognition result determination section This is a section for determining the arrival airport name of the route on which the user wants to know the operation status.

(212) Operation Status Guidance Unit When the route information service is used, the status of the route is spoken.

The above is the processing flow when the route information service is selected.

Next, with reference to the flow charts shown in FIGS. 13 to 16, the processing procedure of the aircraft departure and arrival guidance service in the voice recognition response apparatus according to the present embodiment will be described.

In this processing, first, a service to be used is selected (step 251). It is started by the following system prompt: "This is an aircraft departure and arrival service.""This service provides information on the status of aircraft based on the flight number you want to check or the airport at which the route departs and arrives." Do you use the service? "

In response to this, there is a response from the user. Next, it is checked whether the response from the user is recognizable (step 252).

If the response from the user cannot be recognized (NO in step 252), the flow advances to step 253.
When the recognition is possible (YES in step 252), the recognition result is confirmed (step 254).

Here, the system prompt is "Is it OK with the flight number service?" Or "Is it OK with the route service?"

On the other hand, if the user answers "yes" (YES in step 254), the service to be used is determined (step 255), but if the user answers "no" (NO in step 254). , And proceed to step 253.

In the case of NO in step 252 and in the case of NO in step 254, the process proceeds to step 253, where an inquiry about the service to be used is performed.

Here, the system prompt is “flight name?” Or “route?”.

On the other hand, the user utters a response. Next, it is checked whether the utterance of the response is recognizable (step 256).

Here, if the recognition is not possible (step 25)
(NO at 6), returns to the process at step 253, but if the recognition is possible (YES at step 256), the process of determining the used service is performed (step 255).

That is, if the user answers “Yes” to the question “Flight name?”, The service to be used is determined to be “Flight name”.

If the user answers "No" to the question "Flight name?", The service to be used is determined to be "Route".

If the user answers "yes" to the question "is it a route?", The service to be used is determined to be "route".

If the user answers "No" to the question "is it a route?", The service to be used is determined to be "flight name".

When the service to be used is determined by the above processing,
Next, the utterance amount of the user is checked (step 25).
7). In this case, since the utterance of the user has been acquired several times in the processing so far, the utterance amount of the user is checked based on the acquired data (step 257).

Next, the volume of the system prompt is adjusted based on the user utterance amount obtained in the process of step 257 (step 258).

This is performed based on a system prompt sound volume determination table created based on the user's utterance amount or a system prompt sound volume calculation formula created based on the user's utterance amount. This processing will be described in detail later. .

Next, the type of the guidance service is checked to determine whether it is a flight number service (step 259). This can be determined based on this information because the type information of the used service has already been obtained in the process of step 255.

Here, in the case of the flight number service (YES in step 259), the flow proceeds to step 260 in FIG. 14, and in the case of the route service (NO in step 259), the flow proceeds to FIG.
Proceed to step 270.

First, the flight number service will be described with reference to FIG.

In this process, first, a flight number is obtained (step 260).

The system prompt in this case is "Please tell us the flight number you want to look up."

[0213] On the other hand, it is assumed that the user has made a response of, for example, "ichiichishibin".

Then, it is checked whether or not the utterance of this response can be recognized (step 261).

Here, if the recognition is not possible (step 26)
(1; NO), the process returns to step 260, but if recognition is possible (YES in step 261), a flight number confirmation process is performed (step 262).

By the way, the repetition prompt in this case is
Instead of "Is it a little bottle?", "Is it a little bottle?" A method of creating a repeat prompt in such a case will be described later in detail.

On the other hand, the user utters a response. In this case, there are cases where the utterance of the response can be recognized and cases where the utterance of the response cannot be recognized.
"No" response.

Here, when the response of “Yes” of the user is confirmed (YES in step 262), the flow proceeds to step 263, and the flight number is determined.

Then, the operation status of the corresponding flight is notified (step 264). In this method, the operation status is acquired from a database, and the following guidance information is created, for example, and then spoken to the user. "Every flight left Oki Airport at 15:00 and arrived at Itami Airport at 16:30."

On the other hand, if the response of “No” from the user is confirmed in the process of step 262 and the utterance of the response cannot be recognized, the recognition result is checked, and 1 or 7 is included in the recognition result. It is checked whether or not it is performed (step 265).

Here, if 1 or 7 is not included in the recognition result (NO in step 265), the process returns to the flight number acquisition process in step 260, but if 1 or 7 is included in the recognition result (step 265). Then, the process of creating a pronunciation example prompt of 7 is performed (step 266). this is,
This is because the pronunciation of “Shi” is similar to the pronunciation of “Ichi” and is easily misrecognized.

Then, the following system prompt is created and spoken. "If you pronounce the number" 7, "say" Nana. "

Then, the flow returns to step 260 for obtaining the flight number.

The above is the contents of the flight number service.

Next, the route service will be described with reference to FIG. 15 and FIG.

In this processing, first, the departure airport name is obtained (step 270).

In this case, the system prompt is "Please tell us the departure airport name of the route you want to check."

[0228] On the other hand, for example, if the user
Suppose you respond

Then, it is checked whether or not the utterance of the response can be recognized (step 271).

Here, if the recognition is not possible (step 27)
1 is NO), the process returns to step 270 again, but if recognition is possible (YES in step 271), step 27
Proceeding to step 2, the airport name is confirmed.

In step 270, recognition of the departure airport is obtained based on the response from the user. In this case, a plurality of recognition candidates are obtained in the order of their certainty factors and stored as recognized airport name candidates. (Step 27
4). The recognized airport name candidate stored in step 274 may be used in the process of step 277 as described later.

It is to be noted that what is confirmed in the process of step 272 is the airport name having the highest certainty factor among the recognition candidates recognized as described above.

The repetition prompt in the confirmation process of step 272 is "Itami Airport?"

On the other hand, the user utters a response. In this case, if the user responds “yes” (YES in step 272), the process proceeds to step 273.
Then, the departure airport name is determined.

[0235] On the other hand, if the user says "Itami" but the device side misrecognizes and the repetition prompt is, for example, "Kitami Airport?"
The user's answer is "No" (N in step 272).
O) In this case, a process of acquiring the name of the prefecture is performed (step 275).

In this case, the system prompt is "Please state the name of the prefecture where the departure airport is located."

[0237] In response to this, it is assumed that the user has responded, for example, "Osaka".

Then, it is checked whether or not the utterance of this response can be recognized (step 276).

Here, if the recognition is not possible (step 27
(NO at 6), returns to the process of step 275 again, but if recognition is possible (YES at step 276), step 27
Proceeding to 7, the matching process between the airport name and the prefecture name is performed.

Here, the airport name identification processing is performed by matching the plurality of recognized airport name candidates stored in the processing in step 274 with the prefecture names acquired in the processing in step 275.

For example, if the recognized airport name candidate stored in the processing of step 274 is “damaged”,
If the prefecture name stored in the process of step 5 is “Osaka”, the departure airport name can be specified as “Itami”.

When the departure airport name is specified in this way, the confirmation processing is performed next (step 278).

The system prompt in this case is "Is it OK at Itami Airport in Osaka?"

Here, if the user utters “Yes” (YES in step 278), the flow advances to step 273 to execute a process for determining the departure airport name.

On the other hand, if "No" is uttered (NO in step 278), the process returns to step 275 again.

When the departure airport name is confirmed in this way, the arrival airport name acquisition processing shown in FIG. 16 is performed (step 28).
0).

In this case, the system prompt is "Please state the arrival airport name."

[0248] In response to this, it is assumed that the user makes a response of "Oki", for example.

Then, it is checked whether or not the utterance of this response can be recognized (step 281).

Here, if the recognition is not possible (step 28)
1 is NO), the process returns to step 280 again.
Proceeding to 2, the airport name is confirmed.

In step 280, recognition of the arrival airport is obtained based on the response from the user. In this case, a plurality of recognition candidates are obtained in the order of their certainty factors and stored as recognized airport name candidates. (Step 28
6). The recognized airport name candidate stored in step 286 may be used in the process of step 289 as described later.

It is to be noted that what is confirmed in the process of step 282 is an airport name having the highest certainty factor among the recognition candidates recognized as described above.

The repetition prompt in the confirmation processing of step 282 is "Oki Airport?"

On the other hand, the user utters a response. In this case, if the user responds “yes” (YES in step 282), the process proceeds to step 283.
Then, the arrival airport name is determined.

On the other hand, if the user says "Oki" but the device side misrecognizes and the repetition prompt is, for example, "Iki Airport?", In the process of step 282, the user's answer is "No". (N at step 282)
O) In this case, a process of acquiring the name of the prefecture is performed (step 287).

The system prompt in this case is "Please state the name of the prefecture where the arrival airport is located."

[0257] On the other hand, for example, if the user
Suppose you respond

Then, it is checked whether or not the utterance of the response is recognizable (step 288).

Here, if the recognition is not possible (step 28)
8 is NO), the process returns to step 287 again, but if recognition is possible (YES in step 288), step 28
Proceeding to 9, the matching process of the airport name and the prefecture name is performed.

Then, the airport name specification processing is performed by matching the plurality of recognized airport name candidates stored in the processing in step 286 with the prefecture names obtained in the processing in step 287.

For example, the recognition airport name candidate stored in the processing of step 286 includes “OK”, and step 287
If the name of the prefecture memorized in the processing of
The arrival airport name can be specified as "Oki".

When the arrival airport name is specified in this way, the confirmation processing is performed next (step 290).

The system prompt in this case is "Is it OK at Oki Airport in Shimane?"

Here, if the user utters "Yes" (YES in step 290), the flow advances to step 283 to execute a process for determining the arrival airport name.

On the other hand, if "No" is uttered (NO in step 290), the process returns to step 287 again.

When the departure airport name and the arrival airport name are determined by the above processing, the route name for which information is provided is determined (step 284), and the operation status of the corresponding route is guided (step 284). 285). In this method, the operation status of the route is obtained from a database, the following system prompt is created, and then the user is spoken. "117
The flight departs from Itami Airport at 9:00 and arrives at Oki Airport at 10:05
Arrived in minutes. "Flight 265 departs from Itami Airport at 12:00 and arrives at Oki Airport at 13:05.""Flight 285 departs from Itami Airport at 16:00 and arrives at Oki Airport at 17:05." intend to do something."

The above is the processing procedure of the aircraft departure and arrival guidance service in the voice recognition response device according to the present embodiment.

Next, the contents of the volume adjustment of the system prompt in step 258 of FIG. 13 will be described with reference to FIGS.

As described above, in the present embodiment, the volume of the system prompt is adjusted in accordance with the utterance amount of the user.

In this case, the volume of the system prompt is increased if the user's utterance is large, and the volume of the system prompt is also reduced if the user's utterance is small.

[0271] This is because the utterance amount of the user is small.
Possible causes include: (1) the volume of the system prompt is too loud and the receiver is released, (2) the user is likely to release the receiver due to his habit, and (3) the voice is originally low.

Therefore, by lowering the volume of the system prompt, a situation in which the system prompt is hard to hear is created, and the receiver is made to approach the face. Also, since a human has a characteristic that his voice is louder when the voice heard is low, the volume of the system prompt is also reduced if the user's utterance amount is small by using this characteristic.

On the other hand, the reason why the user's voice volume is large is (1)
Possible causes are that the volume of the system prompt is too low and the receiver is too close to the face, and (2) the voice is already loud.

Therefore, by increasing the volume of the system prompt, it works to make the receiver slightly away from the face.

For this reason, in this embodiment, as shown in FIG. 17, a relational expression between the user's utterance amount and the system prompt volume is prepared, or as shown in FIG. 18, the user's utterance amount and the system prompt volume are prepared. Is prepared in advance.

In FIGS. 17 and 18, the volume of the system prompt is increased when the volume of the user's voice is large, and the volume of the system prompt is also reduced when the volume of the user's voice is small.

Next, a method of creating a repeat prompt in the recognition result confirmation processing in step 262 of FIG. 14 will be described. This is the case when creating a numeric prompt without digits.

In this processing, for example, w
av format, (1) 0. wav (zero) -9. 1 of wav (kyu)
Digit prompt (2) 00. wav (zero), 01. wav (Zeroichi), 02. wav (53) 53. wav (go-san) ... 99. Prepare a total of 110 prompts, two-digit prompts for wav (kyukyu). Then, a repeat prompt is created using the 110 prompts.

Here, first, an even digit (2n digits, n is an integer)
When the repetition prompt is created, the repetition prompt is created using the two-digit prompt of the above (2). That is, a repetition prompt is created using n two-digit prompts.

For example, to create a repeat prompt “345612”, wav, 56. wa
v, and 12. Connect the wavs.

When generating an odd-numbered ((2n + 1) -digit) repetition prompt, n 2-digit prompts in (2) are used and 1-digit prompts in (1) are replaced by 1.
Use this to create a repeat prompt.

For example, to create a repeat prompt “1038859”, wav, 38. w
av, 85. wav, and 9. Connect the wavs.

Here, considering the flight number repeat prompt of the aircraft, the flight number is represented by three digits, so that one-digit and two-digit prompts are used one by one.

For example, to create a repeat prompt of “035”, use 03. wav, 5. Connect the wavs.

When generating an odd-digit repetition prompt, it is necessary to use one one-digit prompt.
In this case, in the above example, for example, “1038859”
10. When creating a repeat prompt, wav
And 38. wav, 85. wav, and 9. The wavs were tied together and the one digit prompt was used for the least significant digit.

However, as another example of use, a one-digit prompt can be used for the most significant digit.

For example, “1038859” similar to the above
When creating a repetition prompt, wav,
03. wav, 88. wav, 59. Connect the wavs.

There is also a usage method in which a one-digit prompt is not used for the most significant digit or the least significant digit.

For example, to create a repeat prompt “1038859” wav, 3. wav
And 88. wav, 59. Connect the wavs.

Next, step 277 in FIG.
The process of matching the airport name and the prefecture name in step 289 will be described. This is because, as mentioned earlier, if the departure airport name or arrival airport name is misrecognized, the candidate airport name that has already been acquired is matched with the newly acquired prefecture name, and the airport that is deemed to be correct Determine the name.

Hereinafter, a case where the departure airport name is erroneously recognized will be described with reference to FIG.

In this process, first, in the departure airport name acquisition process (step 270) of FIG. 15, a recognized airport name candidate is also acquired (step 274). In FIG. 19, reference numeral 291 denotes a recognized airport name candidate acquired at this time. In the recognized airport name candidate 291, “Kitami”, “Itami”, and “Iwami” are stored in a priority order based on the utterance of the user. On the other hand, the apparatus stores an airport and location relationship table for each prefecture. For example, in Hokkaido where Kitami Airport is located, an airport and location relationship table 292 is stored.
However, in Osaka prefecture where Itami Airport is located, table 293
However, a table 294 is stored for Tottori prefecture where Iwami Airport is located.

Here, if there is an erroneous recognition, it is assumed that the user is asked to speak the name of the prefecture where the airport is located (step 275), and there is an utterance 295 "Osaka".

[0294] Then, the table 293 for Osaka is referred to, and the airport name that is on the recognized airport name candidate 291 and registered in the table 293 is searched. That is, the matching process between the airport name and the prefecture name is performed.

As a result, it is found that the airport name that the user has said is Itami Airport.

By the way, as described above, in the dialogue with the user, the utterance of the user may be unrecognizable or erroneous recognition may occur, but such unavailability or erroneous recognition must not be repeated many times. .

[0297] Here, "unrecognizable" is a problem before erroneous recognition, which occurs when the voice of the user cannot be heard or there is an utterance not prepared in the state. In addition, erroneous recognition is likely to occur when there is a problem in how the user speaks.

However, the reason why recognition is not possible or erroneous recognition has occurred is generally not known to the user, so that it is not possible to devise ways of changing the way of saying. Therefore, the same wording is repeated, and unrecognition and erroneous recognition are repeated many times.

Therefore, in the present embodiment, in the case where the recognition cannot be performed the specified number of times or the recognition has been repeated, the utterance of the user who has been used up to that time is used for the repeat prompt, so that the drawback of the user, that is, the voice that is too fast, By making the user aware of small, unclear pronunciation, etc., recognizing failure or erroneous recognition is prevented from being repeated.

FIG. 20 is a flowchart showing a processing procedure in this case, in which the arrival airport name is recognized.

In this process, first, prior to the process of recognizing the arrival airport name, a value n indicating the number of times of non-recognition or the like is initialized (step 300).

Next, it is confirmed that the value of n is smaller than the specified number of times α (step 301), and the value of n is incremented by 1 (step 302).

Subsequently, the arrival airport name is recognized, and the system prompt at this time is "Please state the arrival airport name."

Here, the user's “damaged” or “damaged”
If the utterance is recognized, it can be recognized. Even if "Tami" can be recognized, what can be recognized is that the pronunciation of "i" is a weak pronunciation, and in many cases a recognition trigger (recognition start) is not activated with this word. Therefore, the word “ta” is used as a trigger to recognize “tami”.

On the other hand, the recognition failure occurs when the user is silent or when there is an utterance such as "yes" unrelated to the state at that time.

Next, in step 304, it is checked whether or not recognition is possible. If recognition is not possible, (N in step 304
O), the process returns to the step 301.

On the other hand, if the recognition is possible (YES in step 304), the arrival airport name is confirmed (step 305).

If the data to be recognized is "damaged", the system prompt at this time is (1) "Is it OK at Itami Airport?"

When the recognition target data is "Tami", when "Tami" is recognized as Kitami, (2) "Is it OK at Kitami Airport?"

Here, the user answers "yes" to the system prompt of (1) (step 305).
Then, a recognition result confirmation process is performed (step 306).

Also, since "No" is answered to the system prompt of (2) (N in step 305)
O) Returning to the processing of step 301, if the value of n is less than the specified number of times α, the processing of step 301 and thereafter is performed again.

[0312] In the above processing, the words in "" that the user uttered are stored each time as the contents of the user's actual utterance.

Next, the arrival airport name is not confirmed (step 30).
(NO in 5), and if it is determined in step 301 that the value of n is equal to or more than α, the flow advances to step 308 to perform a process of generating a prompt for returning the user's utterance. This prompt is generated based on the stored contents of the utterance of the user.

Here, the system prompt is as follows.

(1) For example, in the case of silence, "the voice of the customer is too small." (2) In addition, when saying something unrelated, "The customer says" Yes "when talking to the arrival airport." (3) If you misrecognized, "The customer says" Tami "when speaking to the arrival airport."
Becomes

Next, the processing of step 301 and subsequent steps is performed. In the arrival airport name recognition processing of step 303, the user changes the utterance method in response to the system prompt as described above, and utters the following utterance. It is expected to be.

The system prompt of (1) is a little louder.

For the system prompt of the above (2), the name of the existing airport is said.

In response to the system prompt of (3), the word "I" of "Itami" is clearly and aloud.

When the arrival airport name is confirmed by such a newly changed system prompt (step 3)
05 (YES), a recognition result confirmation process is performed (step 3).
06).

By the way, in this kind of apparatus, when the number of data is very large, for example, when recognizing an area code, it is necessary to easily obtain the data originally desired.

Therefore, a method for easily obtaining the data originally desired when the number of data for each category is very large will be described below with reference to an area code.

FIG. 21 is a flowchart showing the processing procedure in this case.

In this process, first, the number n of times of unrecognizable or erroneous recognition is initialized (step 320).

Next, a national area code dictionary is set (step 321).

Here, the area code dictionary will be described with reference to FIG. 22. The area code dictionary includes a nationwide city code in which all the area codes of 2 to 5 digits used nationwide are registered. Area code dictionary 310, Hokkaido area code list dictionary 311 in which area codes are registered in each prefecture, Tokyo area code list dictionary 312, Osaka area code list dictionary 3
13. A prefecture-level dictionary such as the Kagoshima prefecture area code list dictionary 314, and a Hokkaido city in which the area code is registered in association with the municipalities created in each prefecture and included in the prefecture. Municipal name-area code dictionary 315,
There are municipal-level dictionaries such as a Tokyo city / ward / town / area code dictionary 316, an Osaka prefecture / ward / town / village name / area code-related dictionary 317, and a Kagoshima prefecture / city / town name / area code-related dictionary 318. .

Therefore, in the process of step 321, FIG.
Among the area code dictionaries shown in FIG. 2, a national area code dictionary 310 in which area codes of the whole country are registered is set. In the following area code recognition processing, the national area code dictionary 310
Combinations of numbers other than those registered in are not recognized.

Next, area code recognition processing is performed (step 322).

Here, the system prompt is “Please tell us the area code of your telephone number.”
Becomes

On the other hand, there is an utterance of the user's response, and it is checked whether or not the user's response is recognizable (step 323).

Here, for example, if the user's response “Zeroichi Sanroku” is recognized (Y in step 323)
ES), the area code confirmation processing of step 324 is performed, but if the user is silent or there is an utterance "yes" unrelated to this state, the recognition becomes impossible (step 3).
(NO at 23), the process proceeds to step 326.

Next, it is assumed that the system prompt for confirming the area code at step 324 is "Is it all right?"

On the other hand, when the user utters “Yes”, the area code recognition can be confirmed (YES in step 324), and the area code determination processing is performed (step 325). ), End the processing of this state.

On the other hand, it is assumed that the system prompt for confirming the area code in step 324 is "Are you sure you want to get a Zenerana?"

On the other hand, if the user utters “No”, the area code is erroneously recognized (NO in step 324), and the flow advances to step 326.

Step 326 is a step to be performed when there is an unrecognizable or erroneous recognition. Therefore, the number n of unrecognizable or erroneous recognitions is incremented by one.

Next, in order to use the dictionary at the prefectural level or the dictionary at the municipal level shown in FIG. 22, area recognition is performed (step 327).

In this case, the system prompt in the case of the first area recognition, that is, in the case of n = 1, is "Please state the name of the prefecture."

In the case of the second area recognition, that is, in the case of n = 2, the system prompt is "Please state the name of the prefecture and the name of the municipality."

On the other hand, the user utters, for example, "Hokkaido" in the case of the first area recognition, that is, when n = 1.

In the case of the second area recognition, that is, in the case of n = 2, for example, "Kutchan-cho in Hokkaido" is uttered.

Subsequently, the presence or absence of area recognition is checked (step 328). If recognition is not possible (N in step 328)
O), the process returns to step 327, but if recognition is possible (YES in step 328), confirmation processing of region recognition is performed (step 329).

In this case, the system prompt in the case of the first area recognition, that is, in the case of n = 1, is "Is it OK in Hokkaido?"

In the case of the second area recognition, that is, in the case of n = 2, the system prompt is "Is it OK in Kutchan-cho, Hokkaido?"

On the other hand, if the user utters “Yes” (YES in step 329), the process proceeds to step 330, but utters “No” (N in step 329).
O), the process returns to the step 327.

Next, at step 330, it is checked whether or not n = 1, that is, whether it is the first area recognition or the second area recognition. YES), setting of a regional area code dictionary at the prefecture level is performed (step 331). That is, the Hokkaido area code list dictionary 311, the Tokyo area code list dictionary 312, the Osaka area code list dictionary 313, the Osaka area code list dictionary 313 in which the area code is registered for each prefecture in the area code dictionary in FIG. One area / area code list dictionary of prefectures recognized as regions from among the prefecture level dictionaries such as the area code list dictionary 314 is set.

[0347] Then, the area code recognition processing of step 322 and subsequent steps is performed again. In this processing, since the area dictionary divided for each prefecture is used, recognition failure and erroneous recognition are reduced. For example, when the Hokkaido area code list dictionary 311 is used, the first two digits of the Hokkaido area code are “01”, so that only the area code starting with 01 is recognized. Therefore,
Unrecognizable or erroneous recognition is reduced.

On the other hand, in the case of n = 2 and the second region recognition (NO in step 330), a dictionary relating to city / town / village name-area code is set (step 332). That is,
22. The Hokkaido city / city name / area code dictionary 315 created for each prefecture in the area code dictionary in FIG. 22 and registered with the area code in association with the city / town name included in the prefecture. , Tokyo city / ward / town / village name-area code related dictionary 316, Osaka prefecture / ward / town / village name-area code related dictionary 31
7. One of the municipalities-level dictionaries, such as a municipal level-dictionary 318, etc., from the municipal level dictionary, such as the Kagoshima municipal name-area code dictionary 318, is set. .

For example, if the confirmed prefecture name and municipal name are “Kutchan-cho, Hokkaido”, the area code is searched using the dictionary 315 relating to the municipal name-area code of Hokkaido. (Step 333).

That is, a municipal name-area code dictionary 315 associated with a municipal name and an area code of Hokkaido.
Search for Kutchan-cho from, and search for area code. And
The area code is determined (step 3
25).

Therefore, the second embodiment has the following advantages. (1) Since the same system prompt is not repeated as much as possible during a conversation, the user is not discomforted. (2) Relevant questions are asked without repetition of questions of the same content, and the recognition result is narrowed down. The recognition accuracy is increased, so that the conversation time can be reduced. (3) Based on how the user responds, the apparatus works on the user to create an easy-to-recognize situation, and the recognition accuracy is increased, so that the conversation time can be similarly reduced. (4) Since a natural number repetition prompt is created, a low-cost, easy-to-hear system prompt can be created.

By the way, as described above, in the conventional voice recognition response device, there are roughly the following two methods for confirming the utterance of the user.

(1) Items to be confirmed are individually presented,
Ask for the user's answer for each item.

For example, the user answers "Yes" to the system prompt "Is the departure airport Haneda?" Or, the user answers "yes" to the system prompt "Is the arrival airport Itami?"

(2) A plurality of items to be confirmed are presented at a time, and the user's answer is requested for all items.

For example, a user answers "Yes" to the system prompt "Is the departure airport Haneda and the arrival airport Itami?"

However, the conventional checking method as described above has the following problems.

First, in the method of confirming individually as shown in (1), the time of the confirmation work becomes longer and a mechanical dialogue occurs.

In the case of confirming a plurality of items at once as shown in (2), when the user answers "No",
It is difficult to specify which of the plurality of items has been erroneously recognized, and it takes longer time than the case (1).

In the conventional voice recognition response device, when the user responds to the system prompt on the device side for confirmation, if the user is instructed in advance how to respond to the confirmation, the user is generally accepted. Should respond in accordance with the instruction, but there is a problem that some users are confused or missed and do not know what to answer and cannot proceed.

Therefore, as a third embodiment, the state of dialogue between the device and the user is monitored, the question format is changed according to the state of dialogue between the device and the user, and a natural dialogue close to that of a person is realized. The described speech recognition responder will be described.

FIG. 23 is a block diagram showing the overall configuration of the voice recognition / response apparatus according to this embodiment. In FIG. 23, a voice recognition response device 340 includes a voice input device 341,
Voice recognition device 342, recognition word input time monitoring device 343,
Recognition result determination device 344, dialog flow management device 3
45, voice prompt synthesizer 346, voice output device 3
47, a frame information storage unit 348, a dialog flow database (dialog flow DB) 349, a prompt database (prompt DB) 350, and an output prompt storage unit 351.

Here, the voice input device 341 is composed of a microphone or the like of a telephone handset, and converts a voice signal into an electric signal.

[0364] The voice recognition device 342 includes the voice input device 34.
The electric signal obtained in step 1 is converted into character information. The voice recognition processing itself uses a known method, and a detailed description thereof will be omitted.

The recognition word input time monitoring device 343 detects the time when the utterance of the user is recognized. In this embodiment, the utterance time of the user is obtained based on the output start time of the composite prompt output from the device side. .

For example, in FIG. 31, when the user utters “Yes” 461 at the time a1 while “Sapporo departure” a is being uttered, it can be understood that the user uttered “Yes” 461 at the time a1. This processing will be described later in detail.

The recognition result determination device 344 performs the following processing based on the voice recognition recognized by the voice recognition device 342.

(1) The utterance content of the user recognized by the voice recognition device 342 is determined. In this embodiment, "Yes"
Is a positive word, "no" is a negative word,
If it is an airport name or date, it is determined as "related word", if there is no recognition result, "not entered", otherwise, "unrecognizable". Further, the output end point is calculated from the time table of the composite prompt, and if the user's utterance is not within a predetermined time from the output end point, it is determined that “uninput”.

(2) Also, it is determined to which confirmation item the user's utterance was made. In the present embodiment, it is determined from the utterance time of the user detected by the recognition word input time monitoring device 343 and which confirmation item the utterance was made from the time table of the output composite prompt.

Next, the dialog flow management device 345
Controls the entire dialog flow and changes the dialog flow based on the determination result of the recognition result determination device 344. For example, if the determination of the "negative word" is made while outputting the prompt of the arrival airport, the state shifts to the arrival airport confirmation state. Even if a "related word" such as an airport name is recognized, a transition is made to the arrival airport confirmation state. Further, when the determination of “not input” or “unrecognizable” is made after the output of the confirmation prompt, the process shifts to the example prompt.

The voice prompt synthesizing device 346 synthesizes a voice prompt to be output based on the dialog flow selected by the dialog flow management device 345. Also, a correspondence table between the reproduction time of the output prompt and the confirmation item is created at the same time.

The audio output device 347 is constituted by a speaker of a telephone handset, etc., and converts the audio prompt synthesized by the audio prompt synthesizing device 346 into an audio signal.

The frame information storage section 348 holds the determination result determined by the recognition result determination device 344.

The dialog flow database (dialog flow DB) 349 holds a dialog flow required in advance. In this embodiment, a dialog flow for acquiring a departure place, an arrival place, and a date, and a dialog flow management device 345. Holds the dialog flow required by.

The prompt database (prompt D)
B) 350 holds data necessary for synthesizing voice prompts by the voice prompt synthesizer 346. In this embodiment, various airport names and date data,
Holds data required for prompt composition such as commonly used wording.

[0376] The output prompt storage unit 351 holds information on the prompts synthesized by the voice prompt synthesizer 346. In the present embodiment, a confirmation prompt and a time table corresponding to the confirmation prompt are stored.

The above is the configuration. Next, the operation will be described with reference to FIG.
4 to FIG. 27 will be described.

In this process, first, a process of acquiring input information from the user is performed. Specifically, the departure place, the arrival place,
A date is obtained (step 360).

Next, a confirmation prompt is created by the voice prompt synthesizing device 346 based on the acquired input information, and at the same time, a correspondence table between the reproduction time and the confirmation item is created (step 361).

Next, an output request for a confirmation prompt is made (step 362), and the output process is performed (step 363). Then, it is checked whether or not the output of the confirmation prompt has been completed (step 364). If the output of the confirmation prompt has not been completed (N in step 364)
O), proceed to step 370 in FIG.

On the other hand, if the output of the confirmation prompt is completed (YES in step 364), it is next checked whether or not there is an input from the user (step 365). (NO in step 365),
An example prompt is output (step 366), and the process returns to step 365.

On the other hand, if there is an input from the user (YES at step 365), it is checked whether the input is a negative input of "No" (step 367), and "No"
Is negative (YE at step 367)
S), the confirmation prompt is corrected except for the denied items (step 368), and the process returns to step 362.

On the other hand, if it is not a negative input of "No" (NO in step 367), it is checked whether or not it is a positive input of "yes" (step 369). If it is not an affirmative input (step 369). NO), an example prompt is output (step 366), and the process returns to step 365.

On the other hand, if there is an affirmative input of “Yes” (YES in step 369), this processing ends.

Next, referring to FIG. 25, when the output of the confirmation prompt has not been completed (NO in step 364)
Will be described.

In this process, it is checked whether or not there is an input from the user (step 370). If there is no input from the user (NO in step 370), the process returns to step 364 in FIG.

On the other hand, if there is an input from the user (YES at step 370), it is checked whether the input is a negative input of "no" (step 371), and "no"
In the case of a negative input of (YE
S), and proceed to step 380 in FIG.

On the other hand, if it is not a negative input of "No" (NO in step 371), it is checked whether or not a related word such as Iki Airport is input (step 37).
2) In the case of inputting related words (YE in step 372)
S), proceed to step 390 in FIG.

On the other hand, if it is not an input of a related word (NO in step 372), it is checked whether or not the input is affirmative (Yes) (step 373). If the input is affirmative (Yes) (step 373). Then, the affirmed item is checked (step 374), and the process returns to step 364 in FIG.

If the input is not affirmative (NO in step 373), the process returns to step 364 in FIG.

Next, with reference to FIG. 26, a description will be given of a processing procedure in the case where a negative input of “No” is made in step 371 of FIG.

In this case, first, the output of the confirmation prompt is stopped (step 380), and the reconfirmation items determined from the time of the user's utterance are extracted (step 381).
Then, a re-input request prompt is output (step 3
82).

Next, it is checked whether or not there is an input from the user (step 383). If there is no input from the user (NO in step 383), the process returns to step 382. If there is (YES at step 383), a reconfirmation prompt is output (step 384), and a response from the user is waited for (step 3).
85).

If there is no input from the user (NO in step 385), the process returns to step 384, but if there is an input from the user, the process returns to step 368 in FIG.

Next, with reference to FIG. 27, a description will be given of a processing procedure when a related word has been input in step 372 of FIG.

In this process, the output of the confirmation prompt is stopped (step 390), and the reconfirmation item estimated at the time of the utterance of the user is extracted (step 391).

Next, a reconfirmation prompt incorporating a related word is created (step 392), and the created reconfirmation prompt is output (step 393). Then, it waits for a response from the user (step 394).

If there is no input from the user (NO in step 394), the process returns to step 393, but if there is an input from the user, the process returns to step 368 in FIG.

The above is the processing procedure of the present embodiment. Next, the operation between the components in FIG. 23 will be described with reference to FIG.

First, the voice input device 341 converts a voice signal input from a user into an electric signal.

The voice signal converted into the electric signal by the voice input device 341 is input to the voice recognition device 342 and the recognition word input time monitoring device 343, and the recognition word input time monitoring device 343.
Then, the input time point of the utterance input to the voice input device 341 is detected and passed to the voice recognition device 342.

[0402] The voice recognition device 342 converts the voice signal converted into the electric signal into character information, and stores the time table held in the output prompt storage unit 351 and the utterance time detected by the recognized word input time monitoring device 343. To switch the recognition dictionary. In addition, the recognition word input time monitoring device 34
If there is no 3, a timer is needed to measure the elapsed time of the output prompt.

The recognition result determination device 344 uses the time table of the output prompt stored in the output prompt storage unit 351 and the utterance time detected by the recognition word input time monitoring device 343 to determine the item to be uttered. Extract,
The content of the recognition result recognized by the voice recognition device 342 is determined. The result is stored in the frame information storage unit 348 and passed to the dialog flow management device 345.

The recognition result determination device 344 uses the time table of the output prompt stored in the output prompt storage unit 351 and the utterance time data detected by the recognition word input time monitoring device 343 to end the output of the prompt. If the user's utterance is not recognized for a certain period of time from the point in time, the information is stored in the frame information storage unit 348 and passed to the dialog flow management device 345.

[0405] In the dialog flow management device 345, the recognition result determination device determines a dialog flow based on the data output from the 344 and the data stored in the frame information storage unit 348, and a dialog flow database (dialog flow DB) 349. Is output to the voice prompt synthesizing device 348 with reference to the stored data.

The voice prompt synthesizer 348 creates a voice prompt by referring to data stored in the prompt database (prompt DB) 350 based on information obtained from the dialog flow management device 345. At the same time, create a timetable for the created prompt. The created prompt and time table are stored in the output prompt storage unit 351.

[0407] Then, the voice output device 347 outputs the voice prompt generated by the voice prompt synthesizer 348.

The above is the operation between the components in FIG.

Next, a case where the dialog flow is changed based on the utterance time of the user and the content determination information will be described with reference to FIG.

In this process, the user answers "Sapporo" to the system prompt "Please give me the departure place" (Step 420) (Step 421), and the system prompt "Please give me the destination." Step 422)
, The user answers "Iki" (step 42).
3) In response to the system prompt "Please give me the date of departure."
This is the process for changing the dialog flow of the confirmation prompt when "day" is answered (step 425).

In this case, first, the confirmation system prompt is "Is it OK for the flight from Sapporo to Oki Airport on July 7?" (Step 426).

In this case, the dialog flow is changed as follows according to the user's utterance.

(1) In the case where the user utterance is “Iki” (step 427) During or immediately after the output of the confirmation prompt for the destination,
If there is a user utterance of "related words" such as "Iki"
Move to the destination confirmation state.

That is, the system prompt "Is it okay at Iki Airport?" Is spoken (step 43).
2) If there is a positive utterance of "yes" by the user (step 433), the processing up to the destination is terminated. In the case of other utterances, the process returns to step 426.

(2) In the case where the user's utterance is “NO” (step 428) During or immediately after the output of the date confirmation prompt,
If there is a user utterance of "negative word" such as "no",
Move to the date confirmation state.

That is, a system prompt “Please give me the departure date” is spoken (step 434).
If the user speaks “August 7” (step 4
35), a confirmation prompt of "Is August 7 ready?" Is uttered (step 436), and if there is a positive utterance of "Yes" by the user (step 437), the step It returns to the process of 426.

(3) In the case where the user utterance is “Yes” (step 429) If the user utterance of “positive word” such as “Yes” is found after the output of the confirmation prompt, the process proceeds to the next phase. The process proceeds to a dialog flow end prompt (step 438).

(4) In the case where the user utterance is silent (step 430) If the user utterance is not recognized after the output of the confirmation prompt, the process proceeds to a prompt which exemplifies how to reply,
An example prompt such as "Please answer" Yes "or" No "" is spoken (step 439).

(5) When User Speech is Unrecognizable (Step 431) When the user utterance is unrecognizable after outputting the confirmation prompt (other than “positive word”, “negative word”, “uninput”) ,
The process shifts to a prompt that exemplifies how to reply, and utters an example prompt such as “Please answer“ Yes ”or“ No ”” (step 439).

[0420] If the user's utterance cannot be recognized during the output of the confirmation prompt, the dialog flow is not changed.

FIG. 30 shows the same case as FIG. 29, but there is a user utterance of a "positive word" such as "Yes" during the output of the confirmation prompt (step 447), and "No" or the like after the output of the confirmation prompt. If there is an input of "after negative" (step 448), the point at which the "positive word" is recognized or the confirmation item before the utterance of the positive word is omitted from the subsequent prompts.

[0422] That is, in response to the confirmation system prompt "Is it OK for the flight from Sapporo to Oki Airport on July 7?" (Step 426), the user is asked "Yes" immediately after the "Sapporo departure" prompt is output. Is input (Step Uuya), "Departure from Sapporo" before the input of "Yes" is excluded from the subsequent prompts.

Therefore, when there is an input of "after negation" such as "no" after the output of the confirmation prompt (step 44)
8), the system prompt is "Is it OK for the flight to Oki Airport on July 7?" (Step 44)
9).

[0424] On the other hand, if the user utters a "positive word" of "yes" (step 450),
The process moves to the next phase or moves to the end prompt of the dialog flow (step 451).

Next, a processing procedure for changing the dialog flow using the time data of the confirmation prompt and for switching the recognition grammar will be described with reference to FIG.

(1) Confirmation Prompt The confirmation prompt is a prompt for asking the user whether or not the recognition result on the device side is correct. "Are you sure you want to travel from Sapporo to Oki Airport on July 7?" (Step 460).

(2) Time Table The time table is the time data of the synthesized confirmation prompt, and the presentation time is determined for each item or for each part such as wording in correspondence with the confirmation prompt.

For example, in FIG. 31, 2 s (2 seconds) is set in the time table a of the departure place, and 4 s is set in the time table b of the destination.
s, 3s for the date “July 7” c, and 6s for the timetable d.

(3) Speech Recognition Time Using the time table (2), it is possible to know at which point in the time table the utterance of the user was uttered.

For example, in FIG. 31, "Yes" 46 of the user is displayed at the time point a1 of the time table a during which "Sapporo departure" is being uttered.
There is an utterance of “1”, a utterance of “Iki” 462 of the user at the time point b1 of the time table “b” uttering “Oki Airport”, and c1 of a time table “c” uttering “July 7”.
At the time point, the user has an utterance of “different” 463, and the timetable d is uttering “Is it okay with the flight?”
At one point in time, there is an utterance of “Yes” 464 by the user, and it is understood that the user's mute is detected at the time point d2 of the time table d, and that the user cannot be recognized at the time point d3 of the time table d.

(4) Judgment of Recognition Result The recognition result includes “positive words”, “related words”, “negative words”,
There are "Not entered" and "Unrecognizable". From the utterance at the time of recognition of the utterance, a1 in the time table a in which "Sapporo departure" is being uttered
If there is an utterance of “Yes” 461 of the user at the time point, the recognition result becomes “positive word”, and it is recognized that there is an utterance of “Iki” 462 of the user at the time point b1 of the time table “b” uttering “Oki Airport”. The result is "related words" and "July 7"
When there is an utterance of "different" 463 of the user at the time point c1 of the time table c during speech, the recognition result becomes a "negative word", and at the time point d1 of the time table d during the speech time, "Is it okay for convenience?" If there is an utterance of “yes” 464 of the user, the recognition result is “positive word”. Similarly, if the user's mute is detected at the time point d2 of the time table d, the recognition result is “not input”. d3
It can be seen that if the user cannot be recognized at the time, the recognition result is determined to be “unrecognizable”. In addition, during the output of the prompt, by switching the recognition grammar, the "related word" can be determined.

(5) Changing Dialog Flow The dialog flow can be changed depending on where the user's utterance is recognized in the confirmation prompt and the content of recognition.

For example, if there is an utterance of "Yes" 461 of the user at the time a1 of the time table a while "Sapporo departure" is being uttered, the recognition result becomes "Positive word" and the confirmation prompt is continued (step 467).

If the user utters "Iki" 462 at the time point b1 of the time table b during the utterance of "going to Oki Airport", the recognition result becomes "related word", and the process shifts to the destination state (step 468). ).

[0435] If there is an utterance of "different" 463 by the user at the time point c1 of the time table c during the "July 7" c utterance, the recognition result becomes "negative word" and the process shifts to the date state (step 469). ).

Also, at the time point d1 of the time table d in which "Is it okay with the flight", "Yes" of the user is displayed.
If there is an utterance of 464, the recognition result becomes “positive word” and the process shifts to an end prompt (step 470).

If the user's mute is detected at the time point d2 of the time table d, the recognition result becomes "not input" and the process proceeds to an example prompt (step 471).

[0438] Also, if the user's unrecognizable state is detected at the time point d3 of the time table d, the recognition result becomes "unrecognizable" and the process proceeds to the example prompt (step 471).

(6) Switching of Recognition Grammar The recognition grammar used in speech recognition is switched to recognize related words. For example, if the airport name of the prompt is incorrect during or immediately after the output of the confirmation prompt for the departure point, a keyword such as another airport name or departure point may be uttered. In order to recognize the utterance of the user, the recognition grammar is switched according to the location of the output confirmation prompt.

In (6) of FIG. 31, the time table c
, That is, the case where the date is switched.

FIG. 32 is a diagram showing a data format example of frame information stored in the frame information storage unit 348 used for selecting a dialog flow.

The frame information 480 contains the confirmation item 481
, A recognition word 482, and a recognition state 483.

[0443] Here, the confirmation item 481 is an item that requires a user's confirmation judgment, and is an item that must be verified whether it is correct when a user's utterance is recognized.

The recognition word 482 is a character string obtained as a result of speech recognition.

The recognition status 483 indicates the recognition status of the recognition word 482 for the confirmation item 481, and includes “confirmed”, “re-input request”, and “unconfirmed”.

[0446] Here, "confirmed" means that the judgment of the recognition result is "positive word", and "re-input request" means that the judgment of the recognition result is "negative word" or "related word". ,
“Unconfirmed” is a case where the judgment of the recognition result is “uninput” and “unrecognizable”.

(1) For example, when “Sapporo Airport” is being output at the confirmation prompt and the user utters “Yes”, it is “Confirmed”. (2) Also, if the user utters “No” while outputting “Oki Airport” at the confirmation prompt, this is a “re-input request”. (3) If there is no utterance of the user after the output of the confirmation prompt, it is “unconfirmed”.

As described above, in this embodiment, the dialogue state between the device and the user is monitored, the question format is changed according to the dialogue state between the device and the user, and a natural dialogue close to the dialogue with a person is realized. Therefore, the following effects are obtained.

(1) Since the question format is changed according to the situation of the dialog between the device and the user, the time for the confirmation work is shortened. (2) Since the question form is changed according to the dialogue situation between the device and the user, an appropriate voice guide can be provided according to the dialogue situation of the user. (3) Also, since the user can proceed with the work without being confused, the load on the user is reduced.

[0450]

As described above, according to the present invention, in a voice recognition responding apparatus that recognizes and responds to a speech from a partner on a telephone, the utterance state detecting means for detecting the utterance state of the partner, The utterance control means for controlling the utterance to the other party based on the utterance state of the other party detected by the utterance state detection means has the following effects.

(1) The utterance state of the other party, such as the user's skill level and environmental level, is detected, and the voice guide, the dialogue, and the flow of the dialogue are changed, so that the user's skill level, the surrounding environment, and the line are changed. , An appropriate response can be made in accordance with the state of the user, a natural conversation without burdening the user can be realized, and the call time can be shortened. (2) Further, it is possible to provide a conversation device that does not give stress to the user, and it is possible to effectively use the telephone line.

According to the present invention, there is provided a voice recognition response device which recognizes and responds to a speech from a partner by telephone.
Utterance content recognition means for recognizing the utterance content of the other party;
Speech content confirmation means for confirming the speech content of the other party recognized by the speech content recognition means, and speech content change means for changing the speech content to the other party based on the confirmation result of the speech content confirmation means, As a result, the following effects can be obtained.

(1) The same system prompt is not repeated as much as possible during a dialogue, so that the user is not uncomfortable. (2) Relevant questions are asked without repetition of questions of the same content, and the recognition result is narrowed down. The recognition accuracy is increased, so that the conversation time can be reduced. (3) Based on how the user responds, the apparatus works on the user to create an easy-to-recognize situation, and the recognition accuracy is increased, so that the conversation time can be similarly reduced. (4) Since a natural number repetition prompt is created, a low-cost, easy-to-hear system prompt can be created.

Also, in the present invention, in a voice recognition response device that recognizes a speech from a partner on a telephone and responds to confirmation, a confirmation that creates a confirmation prompt by dividing the partner's utterance for each confirmation item. Prompt creation means, utterance time setting means for setting an utterance time for each confirmation item in the confirmation prompt, and when there is an input from the other party during the utterance time set by the utterance time setting means, Utterance content detection means for detecting the utterance content of the other party, and input time detection means for detecting the input time of the input utterance of the other party, and the other party detected by the input time detection means Since the response to the confirmation is made based on the input time of the utterance on the side and the utterance content of the other party detected by the utterance content detection means, the following effects are obtained.

(1) Since the question format is changed according to the situation of the dialog between the device and the user, the time for the confirmation work is shortened. (2) Since the question form is changed according to the dialogue situation between the device and the user, an appropriate voice guide can be provided according to the dialogue situation of the user. (3) Also, since the user can proceed with the work without being confused, the load on the user is reduced.

[Brief description of the drawings]

FIG. 1 is a schematic block diagram showing the overall configuration of a speech recognition response device according to a first embodiment to which the present invention has been applied.

FIG. 2 is an explanatory diagram of frames and semantic information stored in a frame / semantic dictionary shown in FIG. 1;

FIG. 3 is an explanatory diagram of a case where one recognition result is determined from a plurality of recognition result candidates by a recognition result determination device by using the slot constraint shown in FIG. 2;

FIG. 4 is a view for explaining a proficiency level determination processing method determined by the proficiency level determination device shown in FIG. 1;

FIG. 5 is an exemplary view for explaining an environment degree determination processing method determined by the environment degree determination apparatus shown in FIG. 1;

FIG. 6 is an exemplary view for explaining a method of selecting a voice guide selected by the voice guide selecting device shown in FIG. 1;

FIG. 7 is a flowchart showing a processing procedure according to the first embodiment to which the present invention is applied;

FIG. 8 is a view for explaining the operation between the components shown in FIG. 1;

FIG. 9 shows a conventional airline ticket reservation in which the departure point is Haneda, the arrival point is Sapporo, and the departure date is December 31, 1998 at 5:30 pm. Explanatory drawing of an audio guide.

FIG. 10 is an explanatory diagram in the case where there is an erroneous recognition in the voice guide of FIG. 9;

11 is an explanatory view showing a case where there is an erroneous recognition in the voice guide of FIG. 9;

FIG. 12 is a configuration diagram illustrating an overall processing flow according to the second embodiment;

FIG. 13 is a flowchart illustrating a processing procedure of an aircraft arrival / departure guidance service in the voice recognition response device according to the first embodiment to which the present invention is applied.

FIG. 14 is a flowchart illustrating a processing procedure of an aircraft departure and arrival guidance service in the voice recognition response device according to the first embodiment to which the present invention is applied.

FIG. 15 is a flowchart illustrating a processing procedure of an aircraft arrival / departure guidance service in the voice recognition response device according to the first embodiment to which the present invention is applied.

FIG. 16 is a flowchart illustrating a processing procedure of an aircraft arrival / departure guidance service in the voice recognition / response apparatus according to the first embodiment to which the present invention is applied.

FIG. 17 is a diagram showing the relationship between the volume of the user's utterance and the volume of the system prompt, in which the volume of the system prompt is increased if the volume of the user's utterance is large, and the volume of the system prompt is also reduced if the volume of the user's utterance is small; .

FIG. 18 is a table showing the relationship between the volume of a user's utterance and the volume of a system prompt in which the volume of a system prompt is reduced when the volume of the user's voice is large and the volume of the system prompt is also reduced when the volume of the user's voice is small; .

FIG. 19 is an exemplary view for explaining matching processing between an airport name and a prefecture name;

FIG. 20 shows a case where recognizing is impossible or erroneous recognition is repeated a predetermined number of times, and the utterance of the user who has been used so far is used as a repetition prompt. 9 is a flowchart illustrating a processing procedure when the setting is not performed.

FIG. 21 is a flowchart illustrating a method of easily obtaining desired data when the number of data for each category is very large, taking an area code as an example.

FIG. 22 is an explanatory diagram of an area code dictionary. A nationwide area code dictionary in which all area codes of 2 to 5 digits used in the whole country are registered, and an area code of each prefecture. FIG. 4 is an explanatory diagram of a registered prefecture-level dictionary and a dictionary of a municipal level in which an area code is registered in association with a municipal name created for each prefecture and included in the prefecture;

FIG. 23 is a block diagram showing an overall configuration of a speech recognition response device according to a third embodiment.

FIG. 24 is a flowchart illustrating a processing procedure according to a third embodiment to which the present invention is applied.

FIG. 25 is a flowchart showing a processing procedure of a third embodiment to which the present invention is applied.

FIG. 26 is a flowchart showing a processing procedure of the third embodiment to which the present invention is applied.

FIG. 27 is a flowchart showing a processing procedure according to the third embodiment to which the present invention is applied.

FIG. 28 is a view for explaining the operation between the components shown in FIG. 23;

FIG. 29 is an explanatory diagram in the case where the dialog flow is changed based on the utterance time of the user and the content determination information.

FIG. 30 is an explanatory diagram showing a case where the dialog flow is changed based on the utterance time of the user and the content determination information.

FIG. 31 is a flowchart showing a processing procedure when a dialog flow is changed using time data of a confirmation prompt and when a recognition grammar is switched.

FIG. 32 is a diagram showing a data format example of frame information stored in a frame information storage unit 348 used for selecting a dialog flow.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 10 Speech recognition response device 11 Speech input device 12 Speech waveform feature extraction device 13 Speech recognition device 14 Recognition result judgment device 15 Proficiency judgment device 16 Environment judgment device 17 Dialogue control device 18 Voice guide selection device 19 Waveform feature analysis device 20 Speech Output device 21 Voice guide dictionary 22 Recognition dictionary 23 Waveform feature database (Waveform feature DB) 24 Frame / meaning dictionary 25 Recognition result database (Recognition result DB) 26 Dialogue history database (Dialogue history DB) 30 Airline ticket reservation frame 31 Meaning information 31a Airport slot restriction 31b Date slot restriction 31c Time slot restriction 35 User utterance (actual user utterance) 36 Recognition processing 37 Recognition determination processing 38 Application (app body) 40 Rules for voice guidance 41 Voice guide database (voice guide D) B) 201 service selection state 202 selected service confirmation state 203 Yes No state 204 recognition result determination section 205 previous state user utterance amount check section 206 system prompt volume control section 207 service branch section 208 flight number recognition state 209 flight number confirmation state 210 digit pronunciation Law prompt generation unit 211 Recognition result determination unit 212 Operation status guidance unit 213 Departure airport name recognition state 214 Departure airport name confirmation state 215 Prefectural name recognition state 216 Airport name confirmation state 217 Recognition result determination unit 218 Arrival airport name recognition state 219 Arrival Airport Name Confirmation State 220 Prefectural Name Recognition State 221 Airport Name Confirmation State 222 Recognition Result Determining Unit 290 Relation Table 291 between User Speech Volume and System Prompt Volume 291 Knowledge airport name candidate 292, 293, 294 Airport and location relation table 295 Utterance 310 Nationwide area code dictionary 311 Hokkaido area code list dictionary 312 Tokyo area code list dictionary 313 Osaka area code list dictionary 314 Kagoshima area Area Code List Dictionary 315 Hokkaido Municipal Area Name-Area Code Related Dictionary 316 Tokyo Metropolitan Area Municipal Name-Area Area Code Related Dictionary 317 Osaka Prefecture Municipal Area Name-Area Code Related Dictionary 318 Kagoshima Prefecture Municipal Area Name-Area Office number related dictionary 340 Speech recognition response device 341 Speech input device 342 Speech recognition device 343 Recognition word input time monitoring device 344 Recognition result judgment device 345 Dialog flow management device 346 Speech prompt synthesis device 347 Speech output device 348 Frame information storage unit 349 Dialog flow Database (Dialog Flow DB) 3 0 prompt database (Prompt DB) 351 outputs the prompt storage unit 480 frame information 481 Check items 482 recognition word 483 recognizes status

──────────────────────────────────────────────────の Continued on the front page (51) Int.Cl. ⁷ Identification FI FI Theme Court ゛ (Reference) H04M 3/42 P (72) Inventor Koji Omoto 10 Hanazono Dodocho, Ukyo-ku, Kyoto-shi, Kyoto In-company (72) Inventor Hiroshi Nakajima 10-Family Todo-cho, Hanazono, Ukyo-ku, Kyoto, Japan 9A001 HH17 HH18 JJ62 KK56

Claims (34)

[Claims] 1. A speech recognition response device for recognizing and responding to an utterance from a partner on a telephone, comprising: an utterance state detector for detecting the utterance state of the other party; and the other party detected by the utterance state detector. An utterance control means for controlling utterance to the other party based on the utterance state of the speech recognition response device. 2. The utterance state detecting means includes: a proficiency level detecting means for detecting the proficiency level of the utterance of the other party; and an environmental level detecting means for detecting the environmental level of the utterance of the other party. The voice recognition response device according to claim 1, wherein: 3. The voice recognition response device according to claim 1, wherein said utterance control means controls utterance content and utterance speed. 4. The apparatus according to claim 2, wherein said proficiency detecting means is detected based on a time until the other party responds to the utterance from the apparatus and a response time of the other party. A speech recognition response device according to claim 1. 5. The environment level detecting means includes: a strength of environmental noise of a partner, a strength of speech of the partner,
3. An environmental level is detected based on a difference between the utterance strength of the other party and the strength of environmental noise of the other party.
A speech recognition response device according to claim 1. 6. A voice recognition and response device for recognizing and responding to a speech from a partner by telephone, comprising: a voice recognition unit configured to recognize the speech of the partner as character information; and a mastery of the recognized speech of the partner. Proficiency detecting means for detecting the degree of proficiency, environmental level detecting means for detecting the environmental level of the utterance of the other party, and changing the content of utterance from the apparatus side based on the proficiency and environmental level of the utterance of the other party. A speech recognition response device, comprising: utterance content changing means. 7. The speech recognition means includes: a recognition candidate extraction means for extracting a plurality of recognition candidates; and a rule for extracting one fixed candidate from the plurality of recognition candidates extracted by the recognition candidate extraction means. The voice recognition response device according to claim 6, comprising: extracting one fixed candidate from the plurality of recognition candidates using the rule. 8. A voice recognition response method for recognizing and responding to an utterance from a partner on a telephone, comprising: detecting the utterance state of the partner; and detecting the utterance state of the partner based on the detected utterance state of the partner. Controlling speech to the side. A speech recognition response method, comprising: 9. The voice recognition response according to claim 8, wherein the detection of the utterance state includes detecting a proficiency level of the utterance of the other party, and detecting an environmental level of the utterance of the other party. Method. 10. The speech recognition response method according to claim 8, wherein the control of the utterance includes controlling an utterance content and an utterance speed. 11. The method according to claim 9, wherein the detection of the proficiency level is performed based on a time until the other party responds to the utterance from the apparatus side and a response time of the other party. Described voice recognition response method. 12. The method of detecting the degree of environment, comprising: determining the strength of environmental noise of the other party, the strength of speech of the other party,
10. The environment level is detected based on a difference between the utterance strength of the other party and the strength of environmental noise of the other party.
The voice recognition response method described in 1. 13. A voice recognition response method for recognizing and responding to an utterance from the other party by telephone, comprising: a step of recognizing the utterance of the other party as character information; Detecting the environmental level of the utterance of the other party, and changing the content of the utterance from the apparatus based on the proficiency level and the environmental level of the utterance of the other party. Voice recognition response method. 14. The voice recognition includes: a step of extracting a plurality of recognition candidates; and a rule for extracting one fixed candidate from the plurality of extracted recognition candidates. 14. The speech recognition response method according to claim 13, wherein one confirmed candidate is extracted from the candidates using the rule. 15. A speech recognition and response device for recognizing and responding to an utterance from a partner on a telephone, comprising: utterance content recognition means for recognizing the utterance content of the other party; A speech recognition response device comprising: utterance content confirmation means for confirming the utterance content of; and utterance content change means for changing the utterance content to the other party based on the confirmation result of the utterance content confirmation means. 16. The utterance content confirmation unit includes: a recognition target candidate detecting unit that detects a plurality of recognition target candidates; and a related recognition target obtaining unit that obtains a related recognition target including the recognition target. If the utterance content recognition unit cannot confirm the utterance content of the other party, the plurality of recognition target candidates detected by the recognition target candidate detection unit and the recognition target acquired by the related recognition target acquisition unit are included. 16. The speech recognition response device according to claim 15, wherein the utterance content of the other party is specified by matching with the related recognition target. 17. The utterance content confirmation means, when the utterance of the other party has a pronunciation of “Shi” indicating 7
The voice recognition response device according to claim 15, wherein the confirmation is performed by replacing with "Nana". 18. The utterance content confirmation means has utterance storage means for storing the utterance of the other party, and when the utterance content of the other party cannot be confirmed a predetermined number of times or more,
16. The speech recognition response device according to claim 15, wherein the utterance content to be confirmed includes the utterance of the other party stored. 19. A voice recognition response device that utters a voice including a number in response to a voice uttered by the other party, comprising: a one-digit voice prompt storage means for storing a one-digit voice prompt from 0 to 9; 2 for storing two digit voice prompts from 00 to 99
Digit voice prompt storage means, and wherein even-numbered numbers generate a voice utterance using only the two-digit voice prompts stored in the two-digit voice prompt storage means, and odd-numbered numbers indicate the one-digit voice prompts. A speech recognition responder, wherein a speech utterance is created by using only one one-digit speech prompt stored in a storage means. 20. A speech recognition response device for extracting a speech item of a partner from a plurality of items based on a speech from a partner of the telephone, a first registration unit for registering the plurality of items, A second registration unit for dividing and registering a plurality of items registered by the registration unit for each item having a specific attribute and a relevance, and a relevancy extraction unit for extracting the relevance, When the utterance item of the other party cannot be extracted by the first registration means, the relevance extraction means extracts the relevance at a specific attribute including the utterance item, and determines the relevance at this specific attribute. A speech recognition response device for extracting a speech item of the other party from the second registration means based on the second registration means. 21. A voice recognition response device for recognizing and responding to an utterance from a partner on a telephone, comprising: a volume detector for detecting a volume of the partner from the utterance of the partner; Response volume control means for controlling the volume of the response based on the volume of the other party. 22. The voice recognition apparatus according to claim 21, wherein the response volume control means increases the response volume as the volume of the other party increases, and decreases the response volume as the volume of the other party decreases. Answering device. 23. A voice recognition response method for recognizing and responding to an utterance from the other party by telephone, comprising the steps of: recognizing the utterance content of the other party; and confirming the recognized utterance content of the other party. Changing the content of the utterance to the other party based on the result of the confirmation. 24. The confirmation of the utterance content includes a step of detecting a plurality of recognition target candidates and a step of acquiring a related recognition target including the recognition target, and confirming the utterance content of the other party. If not, the plurality of recognition target candidates detected in the step of detecting the recognition target candidate are matched with the related recognition target including the recognition target obtained in the step of obtaining the related recognition target. 24. The voice recognition response method according to claim 23, wherein the utterance content of the other party is specified. 25. The step of confirming the content of the utterance includes the step of: when the utterance of the other party has a pronunciation of “Shi” indicating 7
24. The voice recognition response method according to claim 23, wherein the confirmation is performed by replacing with "Nana". 26. The step of confirming the content of the utterance includes storing the utterance of the other party, and if the utterance content of the other party cannot be confirmed a predetermined number of times or more,
24. The voice recognition response method according to claim 23, wherein the utterance content to be confirmed includes the utterance of the other party stored. 27. A voice recognition response method in which a voice including a number is uttered in response to an utterance from the other party by telephone, wherein a one-digit voice prompt storage means for storing a one-digit voice prompt from 0 to 9; 2 for storing two digit voice prompts from 00 to 99
Digit voice prompt storage means, and wherein even-numbered numbers generate a voice utterance using only the two-digit voice prompts stored in the two-digit voice prompt storage means, and odd-numbered numbers indicate the one-digit voice prompts. A voice recognition response method, wherein a voice utterance is created by using only one one-digit voice prompt stored in a storage unit. 28. A voice recognition response method for extracting an utterance item of the other party from a plurality of items based on an utterance from a caller, a first registration unit for registering the plurality of items, The second step of dividing and registering a plurality of items registered in the registration step of each item with a specific attribute and related
And a relevancy extracting step of extracting the relevancy. If the first registration means cannot extract the utterance item of the other party, the utterance item is included in the relevancy extraction step. A speech recognition response method comprising: extracting a relevance in a specific attribute to be extracted; and extracting an utterance item of the other party from the second registration means based on the relevance in the specific attribute. 29. A voice recognition response method for recognizing and responding to an utterance from the other party by telephone, comprising detecting a volume of the other party from the utterance of the other party, and based on the detected volume of the other party. A voice recognition response method, comprising controlling a response volume. 30. The sound according to claim 29, wherein the control of the volume of the response is such that the response volume increases as the volume of the other party increases, and the response volume decreases as the volume of the other party decreases. Recognition response method. 31. A voice recognition response device for recognizing an utterance from the other party by telephone and responding to the confirmation, comprising: a confirmation prompt creating means for dividing the utterance of the other party for each confirmation item to create a confirmation prompt; An utterance time setting means for setting an utterance time for each confirmation item in the confirmation prompt; and an input from the other party when there is an input from the other party during the utterance time set by the utterance time setting means. Utterance content detection means for detecting the utterance content of the above, and input time point detection means for detecting the input time point of the input utterance of the other party, and the utterance content of the other party detected by the input time point detection means A speech recognition response device, which responds to confirmation based on the input time point and the utterance content of the other party detected by the utterance content detection means. 32. A confirmation prompt change for changing the content of the confirmation prompt based on the input time of the utterance of the other party detected by the input time detecting means and the utterance content of the other party detected by the utterance content detecting means. 32. The voice recognition response device according to claim 31, comprising means. 33. A voice recognition response method for recognizing a speech by recognizing an utterance of the other party by telephone, comprising: creating a confirmation prompt by dividing the utterance of the other party for each confirmation item; Setting the utterance time for each confirmation item in the prompt; and, if there is an input from the other party during the utterance time set in the utterance time setting step, the utterance content of the inputted other party is Detecting the input time point of the input utterance of the other party, and detecting the input time point of the utterance of the other party detected in the detecting step of the input time point and the content of the utterance A response to confirmation based on the content of the utterance of the other party detected in the step of performing voice recognition. 34. The content of the confirmation prompt is changed based on the input time point of the utterance of the other party detected in the step of detecting the input time point and the utterance content of the other party detected in the step of detecting the utterance content. The voice recognition response method according to claim 33, wherein: