JPH01243761A - Interacting type voice responding device - Google Patents

Abstract

PURPOSE:To improve accuracy to decide the silence and the end of a user by deciding the silent condition of the user according to a threshold after a responding message K is sent and executing the decision of an uttering end according to the other threshold when there is uttering. CONSTITUTION:When automatic incoming operation is finished, a control part 2 selects a message number n=1 from a responding message storing part 6 and sends the number to trunk lines L1 and L2. At such a timer, a user message recording part 7 is activated and message recording is started. Then, a threshold storing part 10 selects a silent condition deciding threshold (TNA<1>). The control part 2 substitutes the TNA<1> to a threshold (TNA) and executes the decision of the silent condition. On the other hand, when a user voice is detected before the TNA<1> passes, a uttering end deciding threshold (TED<1>) corresponding to the message number n=1 is extracted from the threshold storing part 10 and the TED<1> is substituted to a threshold TED. Then, a timer count is reset and the decision of the uttering end is executed.

Description

DETAILED DESCRIPTION OF THE INVENTION (Technical field to which the invention pertains) The present invention is an interactive voice response device that sends and processes appropriate response messages in 71 voices in response to voice messages from a user. The present invention relates to a device that determines that a user does not start speaking in response to a response message sent from the device, and determines that the user who has started speaking has finished speaking.

(Prior art) The format in which the device responds point by point to voice input from the user (dialogue format) is similar to when humans talk to each other, so
We believe that this is the best form of man-machine interface.

Taking advantage of this characteristic, devices have been developed that aim to improve the recording rate of messages by applying the dialog format to answering machines, which previously had a low rate of recording messages due to the difficulty of speaking (dialogue type answering machines). Apparatus: Patent Application No. 12052/1982).

In this interactive answering machine, when the user starts speaking, the required action of the machine is to send the first response message after detecting that the user has finished speaking. It is.

Normally, when humans have a conversation, they understand the content of the other person's utterance and determine when the utterance has ended by recognizing the phrase breaks in the content. This requires the machine to have the ability to understand the user's voice in real time, which is difficult to achieve with the current state of voice recognition and natural three-word processing. Therefore, the presence or absence of the user's voice is monitored, and the user remains silent for a certain period of time (utterance end determination threshold:
It is determined that the user's utterance has ended when the utterance continues for more than Tgo).

On the other hand, if the user does not start speaking in response to a response message from the machine, the machine may send a response message with a different expression or take actions such as switching the content of the response message to the next topic. required.

In dialogue between humans, the other person's utterances can be determined based on each other's facial expressions.
It can be determined whether or not to start, but as for the operation of the machine,
Based on the silence determination threshold (TNA), when the user does not start speaking after sending a response message and the silence continues longer than the TNA, it is assumed that the other party does not start speaking.

FIG. 6 shows a flowchart of the conventional interactive answering machine described above. That is, the machine performs a response message sending process (1), and at the same time as sending the response message (2), starts recording the message from the user (3).
, At the same time, the timer count is reset (4), and a message (voice) from the user is detected and judged (5)
.

Then, to determine that the user has not started speaking, the time count value T and the silence determination threshold TNA are compared (6).
If ≧TMA, it is determined that the user will not start speaking (7
), stop recording (8).

In addition, in the judgment of the voice detection result (5), when the user starts speaking, the clock ring 1- remains reset (4'), and the voice detection result is not judged (9). If the utterance state continues (sound is heard), the time count remains reset. If one utterance ends and the state becomes silent, the time count value T and the utterance end determination threshold 1. If 'I''≧T0, it is determined that the user's utterance has ended (11), and the recording is stopped (12).

As described above, it is determined that the user does not start speaking ((
5) to (8)) and the determination ((9) to (12)) that the user who started the voice has finished speaking are performed using the respective time thresholds T, lA, and T0. There is.

Conventionally, a fixed TNA value was used for each device, but in reality it should vary depending on the content of the response message.For example, if the content of the response message is difficult to answer, The user needs a long time to think about what to say before starting to speak, and conversely, if the question is simple, the time required to start speaking is short.

Even if the user starts speaking, if the content of the response message sent is difficult to answer, there is a high possibility that there will be a relatively long period of silence during the utterance because the user will think while speaking. .

On the other hand, the next response message to be sent is, for example, "Yes".
, "Yes", etc., the response message should be sent in a well-timed manner during a short period of silence, such as a breather while the user is speaking, but the next response message to be sent has the effect of changing the topic. If it is,
There were problems such as the need to send a response message after the user's message was completely completed.

(Purpose of the invention) The present invention solves the problems of the conventional device described above.

To provide an interactive voice response with a good human interface by accurately determining whether the user has not started speaking in response to a sent response message, or whether the user who has started voice has finished speaking. The purpose is to provide a device.

(Structure of the Invention) (Characteristics of the Invention and Differences from the Prior Art) In order to achieve the above-mentioned object, the present invention sets the optimum mutenjoan judgment threshold 'rN after sending the response message n to the threshold value storage section.
A” (n is the response message number; 1≦m, where m is the total number of response messages) and an optimal utterance end determination threshold T0° after sending the response message n are stored in advance,
After sending the response message (1≦5m), select the threshold 'rNAK from the threshold storage section, determine whether the user is in an irises state based on this, and the user starts uttering the message. In this case, the threshold value T! is stored in the threshold value storage section. 10K is selected and the end of utterance is determined based on this selection.

The conventional technology has poor interactivity because it uses fixed threshold values T, IA, and T0, which are the criteria for determining when the user is silent or when the user starts and ends speaking, but the present invention actually improves the interactivity. The threshold value "rNAn," corresponding to the content of the response message of
``Prepare rgon and find the best threshold T, ,'', TED
The difference is that K is selected and the accuracy and interactivity are good.

(Embodiment) FIG. 1 is a block diagram of an embodiment of the present invention, in which 1 is an incoming call detection section connected to the station 1iAL unit and L2, 2 is a control section composed of a microcomputer,
Reference numeral 3 denotes a loop opening/closing unit for opening/closing the telephone line and the DC loop, and 4 denotes a communication circuit unit connected to the office lines L□ and L2 via the loop opening/closing unit 3.

Reference numeral 5 denotes a response message sending section connected to the transmitting terminal "r", T2 of the communication circuit section; 6 a response message storage section connected to the response message sending section 5 and storing a plurality of response messages in the order in which they are sent; 7 is a user message recording unit connected to the receiving terminals R□ and R2 of the call rjFl path unit 4;
Reference numeral 8 denotes a voice detection unit which is also connected to the receiving terminals R□ and R2 of the communication circuit unit 4, 9 a timer unit for measuring the continuation of the silence state of the user's voice, and 10 a non-d state determination or the end of utterance. This is a threshold value storage unit that stores threshold values ('rNA, Two) for making a determination.

FIG. 2 also shows the response message storage section 6 in FIG.
FIG. 3 shows an example of the internal configuration of the threshold storage section 10 shown in FIG. 1.

Next, the operation of this embodiment will be explained based on FIG. First, when there is an incoming call, the incoming call detection unit 1 detects this and sends an incoming call δ signal to the control unit 2. When the control section 2 receives this incoming call signal, after a predetermined period of time has elapsed, the control section 2 operates the loop opening/closing section 3 to open a loop, and ends the automatic incoming call operation.

The operation after automatic call reception will be explained using the flowchart shown in FIG.

When the automatic call receiving operation is completed, the control unit 2 receives a response message with message number n=1 (No. 4 (1)) from the response message storage unit 6 (as shown in Figure 2, the content of this message is "Yes, this is oo Shoji"). Selectively (Fig. 4(2)), the response message sending section 5 sends the message to the central office line L2 via the telephone communication circuit section 4 (Fig. 4(3)). At this time, the user message recording section 7 is activated to start recording the message of the user, that is, the caller (Fig. 4 (4)).
, the i' state determination threshold -rNA' is selected from the threshold storage unit 10 (FIG. 4 (5)). After this, the control unit 2 controls the threshold value T
Substitute TNAi for NA and reset the clock count according to the flow ((6) in FIG. 4). L, determine the no-J state ((7) in FIG. 4).

Here, if the user's voice is not detected and the user does not start speaking even after T'llA'', that is, it is determined that the user has fallen into the!! )), it is presumed that the user could not hear the response message from the telephone, so after temporarily stopping the operation of the user message recording unit 7 ((9) in FIG. 4), the response message of n=1 is transmitted again. (Figure 4 (10)).

Furthermore, even if the same response message is sent twice (see Figure 4 (lln), if the user does not start speaking, it is determined that no matter how many times the response message is sent thereafter, the user cannot expect to start speaking. , the content of the response message is switched to the next topic ((12) in FIG. 4).

That is, if the user does not start speaking even after sending the n=1 response message twice, the n=3 response message (from Figure 2, this message is "Excuse me, but which way should I go?"). If the user does not start speaking even after sending the n = 3 response message twice, the n = 4 response message (from Figure 2, this message is ``I am away from home at the moment.''). Please change the topic to "Please complete your request."

However, the reply message “Yes” (n =
2), there is no point in repeating it twice when the user is silent, so if the user does not start speaking after sending the response message, the next response message (n= 3) and change the topic.

On the other hand, if the user's voice is detected before the elapse of %T, A', the utterance end determination threshold T corresponding to the message number n=1 of the response message output from the threshold storage unit IO
To extract 01 (Figure 4 (13)), T to Tl1l, ◎
1 and reset the clock count according to the flow (Fig. 4 (14)) L., Determine the end of utterance (Fig. 4 (14)).
15)).

In this state, if the user's voice continues to be silent for 10 or more times and it is determined that the user's message has ended,
(16)), after stopping the operation of the user message recording section 7 (FIG. 4 (17)), the response message storage section 6
Select n=2 response messages (from FIG. 2, the content of this message is "Yes") from
Then, the signal is sent to the local office line RI, , , L via the communication circuit section 4, and the silent state determination threshold TNA2 is taken out from the threshold storage section 10.

After that, this operation is continued until there are no more response messages (from Figure 2 until mouth = 4) (Figure 4 (18)),
Release the line and end the operation.

FIG. 5 shows an example in which the above operating states are arranged in chronological order based on voices exchanged between the user and the machine.

At this time, each threshold value stored in the threshold value storage unit IO has the following relationship.

(A) Silent state determination threshold TIIA In each scene after sending the first to third response messages (n = 1 to 3), the user responds with "Moshi Moshi", "The name of the person with whom the user has business", You will be asked to speak your "user name." These are content that the user has decided on before making the call, or content that can be uttered naturally based on habit.
It is thought that vocalization begins without requiring particularly long thinking time.

On the other hand, since the fourth response message requests the user to record the business, the user needs to summarize the business in a short time.

Moreover, it is expected that it will take time to compile the matter because the person has to take into account the situation that the person did not know before the call that the person with whom the person has the matter is away. .

From the above, TNA'' (n = 1 to 4) has T
MA” Master T IIA” 4 T NA” < T
It is necessary to satisfy IIA'--(1).

(B) Utterance end determination threshold TgD Since the second response message is a response, it is desirable to send it out in a timely manner with a short silence of the user's voice. For this reason, T and Dl should be set to short values.

On the other hand, after the fourth response message is sent, since the user has to speak while summarizing the matter as described above, there is a high possibility that silence due to thoughts will be included while speaking. After sending the fourth response message, T
Unless o is made sufficiently long, it is not possible to reliably determine that the user has finished speaking.

From the above, T
go'< T go” 'F T go3 <'r
go' ------------- (2) must be satisfied.

(Effects of the Invention) Since the present invention is configured as described above, in the interactive voice response device, the user does not start speaking in response to the sent response message, or -
Since it is possible to accurately determine that the user who has started speaking has finished speaking, it is possible to realize an interactive voice response device with a good man-machine interface.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an embodiment of the present invention. 2 is an example of the internal configuration of the response message storage section 6 of FIG. 1, FIG. 3 is an example of the internal configuration of the threshold value storage section 10 of FIG. 1, and FIG. FIG. 5 is an example of the dialogue that takes place between the machine and the user over time, and FIG. 6 is a flowchart showing the determination procedure of a conventional interactive answering machine. DESCRIPTION OF SYMBOLS 1... Incoming call detection unit, 2... Control unit, 3... Loop opening/closing unit, 4... Call circuit unit, 5... Response message sending unit, 6... Response message storage unit, 7 ...User message recording department, 8
- Voice detection unit, 9... Time measurement unit, 10... Threshold value storage unit. Patent applicant Nippon Telegraph and Telephone Corporation Figure 3 Figure 5 AnQ-

Claims (1)

[Claims] After sending the response message n to the threshold storage unit, set the optimum silent state determination threshold T_H_A^n (n is the response message number: 1≦m, where m is the total number of response messages) and after sending the response message n, Optimal utterance end determination threshold T_
E_D^n is stored in advance, and the response message K(1
≦K≦m), the threshold value T_
N_A^K is selected, and based on this, the user's silent state is determined. When the user starts uttering a message, the threshold value T_E_D is selected from the threshold value storage section.
An interactive voice response device characterized by selecting ^K and determining the end of utterance based on this.