JPH02149900A - Voice recognizing and answering device - Google Patents

Abstract

PURPOSE:To improve service to users by providing a continuous voicing detecting means which detects user's utterance and sends out an utterance indication signal for urging a user to correct the voicing utterance to the user in response to the detection. CONSTITUTION:A voice section detection part 123 detects the length of a voice from all input voice signals inputted to a voice recognition part 16D regardless of whether the user voices a speech continuously or not and an excessive utterance length detection part 127 judges whether the detected utterance length is excessive utterance length or not and outputs an excessive utterance length signal when so. A continuous utterance detection part 151, on the other hand, judges whether or not the voice is uttered continuously and outputs a candidate signal for continuous utterance when it is judged that the voice is uttered continuously. Then only when the excessive utterance length signal and candidate signal are inputted at the same time, the continuous utterance is securely decided and the utterance indication signal which urges the user to correct the voicing utterance is sent out of a continuous decision part 152 and, therefore, a continuous utterance means 150 immediately. Consequently, the service to users is improved.

Description

[Detailed Description of the Invention] (Industrial Field of Application) This invention is a voice recognition response device, especially a voice recognition response device that reduces misrecognition caused by poor vocalization of the user!i
il: related.

(Prior Art) Currently, voice recognition and response devices using a single telephone line have been introduced in the financial industry, distribution channels, etc., and services such as balance inquiries and order entry are provided to users. In other words, these services are performed by the user inputting a personal identification number, membership number, service code, etc. by voice from a telephone or the like.

Prior to explaining the present invention, a conventional voice recognition response device will be briefly explained.

FIG. 2 is an explanatory diagram of a conventional voice recognition response device, showing each component as a block diagram.

In the figure, reference numeral 10 denotes a voice recognition response device, and the voice uttered by the user into the transmitter of the telephone that constitutes the external voice input section 20 is supplied via the exchange 30 as an input voice signal. 40 is a host computer constituting the external processing device M, which controls each component of the voice recognition response device M10 in response to an input voice signal and a signal output from the voice recognition response device 10, and controls other external processing devices. It has the function of outputting control signals to equipment.

This voice recognition response device 110 is connected to the exchange 30 and has a line control unit 11 that controls the interface with public lines and the like.
0, a voice recognition unit 1 that recognizes the voice uttered by the user based on the input voice signal sent from the line control unit 110;
20. A voice recognition response device in response to the output from the voice recognition unit 120 and/or the output from the host computer 40? lll0
A main control unit 130 and a main control unit 13 that control each component of
Speech synthesis unit 14 that synthesizes and outputs speech according to instructions from 0
It is 0.

In addition, the voice recognition unit 120 includes a signal sound generation unit 121 for generating a signal sound from the receiver of the telephone 20 to prompt the user to speak, and analyzes the input voice uttered by the user according to this signal sound. A voice analysis section 122 extracts characteristic parameters, a voice section detection section 123 extracts an input voice part and determines a voice section (voice length), and stores characteristic parameters, which are standard bangs of each voice, in a readable manner. A speech dictionary section 124, a matching calculation section 125 that matches the obtained feature parameters of the speech section with each feature variation stored in the speech dictionary 124 and calculates the distance M (dissimilarity) between the two; A determination unit 126 and a speech section detection unit 1 that determine the category with the smallest distance M value as a recognition result based on the distance value for each category obtained in
The input voice length detected in step 23 is set to a predetermined threshold Tth (the maximum voice length that can be correctly recognized by uttering one character or one digit in the case of a number, herein referred to as the utterance overlength value). ), the configuration includes a utterance appropriate length detecting section 127 that detects the utterance as an appropriate utterance length when the length exceeds . Since the configuration of the voice recognition response device 10 is a conventional configuration, further detailed explanation will be omitted.

Next, an example of the normal operation of the conventional voice recognition response device shown in FIG. 2 will be briefly explained with reference to the appendix and (2). As can be understood from Attached Table 1, after snowmaking starts, the voice recognition response device? 1l
The voice synthesis unit 140 is started through the main control unit 130 of the IO, and from this, a synthetic voice instruction (guidance) such as “Please give me your PIN” is issued from the receiver of the telephone 20 via the line control unit 110 and the exchange 30. , the voice uttered by the user according to this guidance is input as an input voice signal from the transmitter to the voice recognition unit 120 through the exchange 30 and the line control unit 110, and the voice is recognized by the voice recognition unit 120, and the recognition result is The recognition result is sent to the host computer 40 through the main control unit 130 and used as a signal for executing necessary external processing.

By the way, when performing voice recognition, after the signal sound generator 121 sends a signal sound "Hira" to the handset of the telephone 20, the voice recognition operation is started, and the input voice is uttered as one character for each signal sound. 1 for each voice, therefore digits
Recognizes each digit.

However, before inputting voice, it sends out a guidance that says, ``In the future input, please separate each character (or one digit in the case of numbers) according to the beep sound.'' However, there are often cases where the user utters the words in succession due to carelessness. Appendix II shows examples of such malfunctions.
This problem is particularly likely to occur when the user is using Honsoukan 10 for the first time. If the length of the input voice is too long, there is a high possibility that it will be misrecognized. It is configured to operate so as to prompt the user to repeat the voice with guidance such as , please input again ~ j, etc.

(Problem to be solved by the invention) However, the conventional device does not detect the appropriate length of utterance (Unono), but does not determine whether it is caused by continuous utterance, and therefore, The problem is that users (well, they don't understand that continuous utterance is a problem) often repeat continuous utterance again, leading to a decline in service to users, such as interrupting operation from the equipment side. But 1.:.

Purpose 1 of the present invention: In the case where the user's carelessness causes continuous voice input, the user can correct the problem with the voice output before it finally ends up interrupting the operation. By maintaining the normal operation of the voice recognition response device in the sliding door, the purpose of the voice recognition system is to improve the quality of service provided to users in each ward. 4- Providing a response device (t).

(Means for Solving the Problems) Aiming to achieve this object 1. According to the voice recognition response device of the present invention, in addition to the conventional functions, it detects that the user has made continuous utterances.] In addition, in response to this detection, the continuous utterance detection means is provided which sends out a 3'' utterance instruction signal to the user (to prompt the user to correct the utterance).

In carrying out the present invention, it is preferable that the consecutive utterance means is configured by a continuous utterance detection section and a continuous utterance determination section. Part 69
Preferably, the continuous utterance detection section is configured to output a candidate signal indicating continuous utterance when the input voice & is a predetermined continuous utterance detection means. or,
A continuous utterance determination section is provided in the main control section (vl), and when the voice ease of the input voice is equal to or greater than a predetermined value of 15 utterance excessive lengths, the continuous utterance determination section is configured to control the utterance appropriate length output from the voice recognition section. Preferably, the arrangement is such that an AND signal is generated between the signal and the aforementioned candidate signal.

(Operation) According to such configuration 15-, regardless of whether the user continuously utters () or not, the voice section detection section detects the voice section from the input voice signal of 1 and all that is input to the voice recognition section of the voice recognition response device. On the one hand, the detected voice length is determined in the appropriate utterance length detecting section as to whether or not it is the appropriate utterance length. A signal is output, and on the other hand, a rapid continuous vocalization detection unit that constitutes a part of the continuous vocalization means tentatively determines whether continuous vocalization is continuous vocalization or apricot. A continuous utterance candidate signal is outputted.Then, the continuous utterance determination section 33, which is provided in the main control section and constitutes the remaining part of the continuous utterance means, outputs the continuous utterance candidate signal and the candidate signal as described above. In this case, the user intentionally or unconsciously extends the length of the voice and utters 1 note. The continuous determining section, ie, the continuous utterance means, immediately sends out a utterance instruction signal urging the user to correct the voice utterance.

Therefore, the continuous utterance detection means detects the user's utterances.1.
:When it is determined that the voice is continuous voice, in response to this determination, it immediately notifies you that it is continuous voice, prompts you to correct the voice method, and sends out 1-Geigns, so that less voice input operations are required. It becomes possible to continue the process smoothly,
Therefore, the service provided to each user can be improved. (Embodiment) An embodiment of the voice recognition response device of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram for explaining one embodiment of the voice recognition response device of the present invention, and in the same figure, the same components as those shown in FIG. special ft
Components are designated by the same reference numerals except when mentioned, and detailed explanation thereof will be omitted.

According to this invention, the continuous utterance detection means 150 is provided in the voice recognition response device. Here, in the voice recognition response device of the present invention, this continuous utterance detection means 150 shown as 100 detects that the user has made continuous utterances, and
The continuous utterance detecting means 150 may have any configuration as long as it has a function of sending a utterance instruction signal to prompt the user to correct the voice in response to detection of continuous utterances, but preferably this continuous utterance detection means 150 It is preferable to configure the continuous utterance detection section 151 and the continuous utterance determination section 152, and in this embodiment, the continuous utterance detection section 151 is added to the conventional components in the speech recognition section. Therefore, in the embodiment of the present invention, this speech recognition section is designated by the reference numeral 160. Further, since this continuous utterance determination section 152 is provided in the main control section in addition to its conventional components (not shown), in the embodiment of the present invention, this main control section is designated by reference numeral 1.
70 is indicated.

This continuous utterance detection section 151 is activated when the speech length detected from the input speech signal by the speech section detection section 123 is determined to be greater than or equal to a predetermined threshold Tthc (hereinafter, this threshold is referred to as a continuous utterance threshold). , - For example, the input speech length and the threshold value T are set so as to output a candidate signal indicating that the speech is discontinuous. It can be configured to determine which comparison is to be performed.

The threshold value vthc in this case is given by the following equation.

Tthc=nXtA +*++++++++++++ (1
) Here, n is the number of characters to be input, for example, in the case of numbers, it is the number of digits, and is a value determined depending on the purpose of use of the device 100. It is stored in any suitable memory (not shown) provided at 170 and read out and provided as required. In addition, tA is the statistically obtained average minimum voice length for each character (each digit).
The minimum voice length is stored in advance in any suitable memory for each voice, and the minimum voice length is also read out and given as required.

Therefore, when using the wooden M100 to verify a "PIN", normally n is 5 digits, and 0,...9
A threshold @Tthc for each digit is calculated in advance from the minimum voice length of each digit up to It can be readably stored in any suitable memory.

On the other hand, the continuous utterance determination unit 152
The length of the utterance corresponding to the input audio signal is determined to be the appropriate utterance length and is output from the appropriate utterance length signal and the candidate signal from the continuous utterance detection unit 151 described above. is a determination unit that reliably determines that the user's voice utterance is not an appropriate length of voice utterance due to prolongation, but is an appropriate length of voice utterance due to continuous voice utterance. Preferably, the continuous utterance determination section 152 is configured with an AND circuit that outputs an AND between the appropriate utterance length signal and the candidate signal. By this,
Only when both signals are input, it is possible to obtain an AND signal that provides a reliable determination result that the utterance is continuous.

This AND signal is outputted to the speech synthesis section 140 as a speech instruction signal that prompts correction of the speech utterance, and in response to this signal, the speech synthesis section 140 communicates to the user through the receiver of the telephone 20 that the current utterance is a continuous utterance. It is possible to notify the user that the message was uttered and provide guidance for re-voicing.

Next, an example of the operation of the voice recognition response device of the present invention shown in FIG. 1 will be explained with reference to Appendix (2). In this embodiment, an example in which a four-digit password is uttered will be explained.

When the wooden M100 is set to the operating state, (1
) The continuous utterance threshold 'Tthc obtained according to the formula is stored, for example, in a memory (not shown) provided in the continuous utterance detection section 151.

First, according to a command from the main control section 170, the speech synthesis section 1
40 to the handset of the telephone 20, and the message "
From now on, please enter each digit one by one according to the beep sound.'' Next, the user instructs the user to ``please enter the PIN number,'' and causes the signal sound generator 121 to generate a ``beep'' sound. In response, the user
When the user continuously utters "Hachi Nana Yon Zero Go," an input voice signal is input from the transmitter of the telephone 20 to the voice recognition response device 100, and the voice section detection unit 123 detects the length of the voice. This voice length is determined by thresholds @T and Tth in the appropriate voice length detection unit 127 and continuous voice detection unit 151.
Compared with C, if L≧7th, the appropriate length of utterance detection unit 127
If L≧Tthc, candidate signals are generated from the continuous utterance detection section 151 and input to the continuous utterance determination section 152, respectively. When both of these signals are detected and an AND signal of both signals is generated in the continuous utterance determination section 152, this AND signal indicates that it has been determined that the user's utterance this time was continuous utterance. Therefore, the main control section 170 sends a command to the speech synthesis section 140 by directly using this AND signal or by replacing it with any other suitable signal, and the speech synthesis section 140 sends a command to the telephone receiver saying, Therefore, it cannot be confirmed.
Following the signal sound of ``dotsu'' like ``2'', guidance such as ``Please say digit by digit.

The user follows this vocal instruction and, following the "beep" sound from the signal image generating unit +21, says "Hachi", "Pip "Nana", "Pip "Yon", "Dotsu "Go", etc. Correctly say each 4-digit number after the "beep" sound. When such a correct utterance is made, the wooden word 100 operates in the same manner as before and sends a message to the user saying, ``Your password is 8.7.4.0.5.'' After this message, the user hears a beep and then responds with ``Yes 1''.

(Effects of the Invention) As is clear from the above description, according to the voice recognition response device of the present invention, when continuous utterances are made due to the user's carelessness, by detecting this, the utterances can be immediately stopped. Since it is possible to notify the user of errors and corrections, the ease of use of the voice recognition response device is improved, and it is possible to provide a voice recognition response device that can improve the service provided to the user.

Separate table■ Example of normal operation by voice recognition response device Appendix I+ Examples of malfunctions in voice recognition response devices

[Brief explanation of the drawing]

FIG. 1 is a block diagram for explaining an embodiment of the voice recognition response device of the present invention, and FIG. 2 is a block diagram for explaining a conventional voice recognition response device. 20... External audio input section (for example, telephone) 30...
Switching machine 40...external processing bag = (for example, host computer) 100
...Voice recognition response bag 110...Line control section, 121-...Signal sound generation section 122...Speech analysis section, 123...Voice section detection section 124...Voice group M section, 125 - Verification calculation unit 126... Judgment unit, 127... Overvocalization detection unit 140... Speech synthesis unit, 150...
Continuous utterance detection means 151...Continuous utterance detection section, 152
... Continuous utterance determination section 160 ... Speech recognition section, 1
70... Main control unit. Patent Applicant Oki Electric Industry Co., Ltd. Agent Patent Attorney Takashi Ogaki

Claims (2)

[Claims] (1) a voice recognition unit that recognizes the input voice from an external voice input unit word by word; a voice synthesis unit that outputs synthesized voice based on voice data for instructing the user to speak;
The main control unit is equipped with a main control unit, and prior to voice recognition, it sends out a signal sound to prompt the user to speak, recognizes the voice uttered character by character for each signal sound, and outputs the recognition result to an external processing device. A voice recognition response device configured to detect continuous utterances by a user and, in response to the detection, send a utterance instruction signal to the user to prompt the user to correct the utterances. A voice recognition response device comprising a detection means. (2) In the voice recognition response device according to claim 1, the continuous utterance detection means is provided in the voice recognition section and indicates that the continuous utterance is continuous utterance when the voice length of the input voice is equal to or greater than a predetermined continuous utterance threshold. a continuous utterance detection section that outputs a candidate signal of utterance, and an overlength utterance signal that is provided in the main control section and is output from the speech recognition section when the voice length of the input speech is equal to or greater than a predetermined utterance overlength threshold; , and a continuous utterance determination unit that generates an AND signal with the candidate signal.