JPH0588480B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION <Field of Industrial Application> The present invention relates to a voice input/output device that performs voice input by voice recognition and voice output by voice response to a large number of user lines.

<Prior Art> A voice input/output device that performs voice input through voice recognition and voice output through voice response is used in a system for making inquiries from users via telephone or the like. In this inquiry system, as shown in Table 1, the system automatically identifies the content of the inquiry based on voice input from the user, and notifies the user of the answer to the inquiry through voice output. In such an inquiry system, it is necessary to simultaneously provide voice input and voice output services to a large number of users.

Table 1 User: Calling the system System: “This is a telephone service. Please enter your service code. P” User: First system: “P” User: Nice System: “Inquiring about your balance. Your account number. User: Ichi System: “P” User: Ni System: “P” User: Sun System: “Your account number is 123. P” User: Yes System: “Your The balance is 12,300 yen. Thank you very much.” Conventionally, voice input/output devices for a large number of user lines have a recognition unit that can simultaneously process m voice inputs, the same number as the number of user lines, and a It used a response unit that could simultaneously process audio output. However, the recognition section has more complicated processing than the response section and requires a larger amount of memory, resulting in a larger device.

On the other hand, Japanese Patent Application Laid-open No. 131250/1983 describes a voice input/output device in which a recognition section is made small. The outline of this device is as follows. Although the user uses voice input and voice output, the time during which voice output is performed is idle time for the recognition unit without voice input. In order to utilize this free time for voice input processing on other user lines, a changeover switch is provided between the user line and the recognition unit, and only the user line with a voice input request is connected to the recognition unit. However, since the number of simultaneous processing lines of the recognition unit is n, which is smaller than m, it is necessary to control so that the number of voice input requests does not exceed n at the same time. Therefore, only when there is a free line for the recognition unit, or in other words, if there is no free line, the system waits until a free line becomes available, and then outputs a voice prompting the user to utter the voice input (voice input prompt sound). output) and accept voice input from the user. As a result, it has become possible to perform voice input and voice output for m users using a recognition unit that simultaneously processes n voice inputs smaller than m.

However, in the multi-line audio input/output device, the user must perform voice input after listening to the voice input prompting sound. For this reason, unskilled users or those who are unfamiliar with voice input/output devices are likely to impatiently wait for a signal and start speaking. Since the recognition process has not started for the voice uttered before the signal, the beginning of the word may be dropped, resulting in erroneous recognition, or the voice may not be processed at all and the input may not be accepted.

A method for preventing erroneous recognition of voices uttered without waiting for the signal or non-acceptance of input is described in Japanese Patent Application No. 1982-44741 filed by the same applicant as the present application. The voice input device according to the present invention has an input command unit that inputs a voice detection signal and a voice input request and outputs a recognition start signal and a voice input promotion request, and when the voice detection signal is output when the voice input promotion request is made. When the end of the voice is detected, a voice input promotion request is output again to restart the recognition unit.
In other words, if you can't wait for the cue and speak, the voice detection signal is output when the voice input promotion request is made, and when the end of the voice is detected, the input command unit issues the voice input promotion request and recognition start signal again. Voices that are uttered incorrectly are ignored, and the next voice that is uttered is recognized. This can prevent erroneous recognition of voice uttered without waiting or non-acceptance of input.

However, the voice input device according to Japanese Patent Application No. 58-44741 is a voice input device for one line;
- When used as a recognition unit for a multi-line audio input/output device using the 131250, the user's line is switched by the switch, so the voice of the user's line cannot be constantly monitored, and the voice uttered by impatient waiting for a signal cannot be accurately detected. It is not possible to prevent erroneous recognition or non-acceptance of input. For this reason, a special request for
A switch cannot be used for a multi-line audio input/output device that has an audio input section based on 44741.
As expected, a recognition section for recognizing m voices is required, which results in a large device.

<Objective of the Invention> The object of the present invention is to provide a voice input/output device which is provided with a changeover switch before the voice recognition processing section, uses the voice recognition processing section in a time division multiplex manner, and notifies the user of the timing of use by voice output. By having a detection unit that detects the voice uttered by the user impatiently waiting for a cue, and an input command unit that outputs a voice input promotion request again when the voice is detected, the user does not have to be so sensitive to synchronize with the cue. To provide a small-sized voice input/output device that can perform easy-to-use voice input for a large number of users.

<Structure of the Invention> A multi-line audio input/output device according to the present invention includes m (positive integer) user lines, a response unit that outputs audio to the m user lines, and a response unit that outputs audio to the m user lines. a detection unit that detects audio input from a user line; and a detection unit that detects the audio input from the
(a positive integer) recognition unit; a recognition unit that recognizes the audio of the n input lines that receive the output of the changeover switch; a control unit that connects a user line that is only in a talking state to an idle line of the recognition unit and outputs a voice input request, and a voice is not detected by the detection unit before the time when the voice input request is received; When a voice is output to the user line to prompt the user to speak, the recognition unit is commanded to start, and when voice is detected, at the time when the end of the voice is detected, a voice is output to the user line to prompt the user to voice. It has an input command unit that instructs the recognition unit to perform the following steps and start the recognition unit.

<Operation/Principle of the Present Invention> The multi-line audio input/output device according to the present invention performs audio input and output to m user lines in accordance with audio input commands and audio output commands from a host system. The specific operation will be explained with reference to FIG. The control unit, which first receives the voice output command, tells the response unit, “Please give me your account number.”
This command instructs the user line m to output the voice. When the audio output ends, the host system is notified of the end via the control unit. The higher-level system that received the notification outputs a voice input command to the control unit. The control unit searches for a free input line of the recognition unit,
When found, a changeover switch connects the requested user line to the input line of the vacant recognition section. Furthermore, it notifies the input command section of the voice input request. Upon receiving the voice input request, the input command unit determines whether voice has already been detected, and if no voice has been detected, instructs the response unit to output a voice input prompting sound (for example, "beep"), and Instructs the recognition unit to start recognition processing for the input line. if,
If the voice from the user's line was detected before the time when the voice input request was received, it means that the user had uttered the voice before the voice input prompt sound, so At the time when the end of the voice is detected, the response unit is again instructed to output the voice input promotion sound, and the recognition unit is instructed to restart the recognition process.

<Example> The present invention will be described in more detail based on a block diagram showing an example shown in FIG. As shown in FIG. 2, the embodiment of the multi-line audio input/output device according to the present invention has user lines 11, 12, ..., 1m,
Recognition unit input lines 21, 22, ..., 2n, voice detection unit 3, changeover switch 4, recognition unit 5, response unit 6,
It is composed of an input command section 7 and a control section 8.

The changeover switch 4 can take the configuration shown in, for example, FIG. 4, FIG. 5, or FIG. 7 of JP-A-56-131250. That is, the changeover switch 4 inputs the user lines 11, 12, .
..., 2n as outputs, and under the control of the control unit 8, only the line requesting voice input from among the user lines 11, 12,..., 1m is selected as the recognition unit input line 2.
Connect to 1, 22,..., 2n. The response section 6 is, for example, the publication IE Transactions on APS (IEEE
Transactions on ASSP), November 1974 issue.
The paper from pages 339 to 352, ``A Multiline Computer Voice Response System Utilizing ADM Coded Speech (a Multiline
Computer Voice Response System Utilizing
The configuration shown in FIG. 2 on page 340 of ``ADPCM Coded Speech'' can be used.
That is, the response unit 6 follows instructions from the control unit 8 or the input command unit 7 to send messages to the m user lines saying, ``This is the telephone center. Please give me the service code.''
It outputs the sentence ``Beep'' and the voice input promotion sound ``Beep''. Speech recognition devices can be used, for example, in the publication PROCEEDINGS OF THE IEEE.
From pages 405 to 415 of the April 1976 issue of
Practical Applications of Voice Input to Machines
The structure shown in FIG. 5 of ``Machines'' can be used. That is, as shown in FIG. 3, there is a pre-processing section that analyzes speech and extracts, for example, power information and feature vectors (spectrum information), a speech detection section that detects at what time the speech was uttered, and The system consists of a feature extracting section that extracts the features of the voice uttered, and an identifying section that identifies the type of voice that was uttered based on the features.

In the present invention, as shown in FIG. 2, the preprocessing section and the voice detection section are separated from the feature extraction section and the identification section, and a changeover switch 4 is placed between them. That is, the voice of the user line i is input to the input line 1i of the voice detection section 3, the time at which the voice is uttered is detected, and the result is sent to the input command section 7. Furthermore, the voice of the user line i passes through the changeover switch 4 and is input to the recognition unit input line 2j. The recognition unit can perform feature extraction and identification processing for n voices.
The voice input from the recognition unit input line 2j is extracted and identified, and the results are output to the control unit 8.

As shown in FIG. 4, the voice detecting section 3 analyzes the voices of m user lines using a preprocessing section 31, and outputs the analysis results to the changeover switch 4. Furthermore, the voice power information of m user lines is sent from the preprocessing section 31 to the power information input section 321 of the microprocessor section 32. The microprocessor section 32 includes an input section 321, a detection information output section 322,
Consisting of a CPU 323 and a memory 324, the fifth
The start and end times of m voices are determined according to the flowchart shown in the figure. In other words, when detecting the beginning of a speech, the assonance level is determined as shown in Figure 5a.
After the time t _s exceeding Th1, the vowel level Th2
If it exceeds , t _s is taken as the starting point time. on the other hand,
As shown in FIG. 5b, in detecting the end of speech, t _e is determined as the end time when the level is below LE time Th1 after the time t _e when the consonant level Th1 has been lowered. The above-mentioned starting end time t _s and ending end time _te are notified from the detection information output section 322 to the input command section 7 . Here, FLAG in FIG. 5 is set to 1 at regular time intervals, and audio is detected from the power Pi at regular time intervals. Also, Di is in a power state,
Di=0, 1, and 2 are states in which there is no sound, a starting edge candidate has been detected, and a vowel has been detected and the starting edge has been determined, respectively. An example of the operation of the voice detection section described above is shown in FIG. 6. In FIG. 5, Pi, Si, and Ei indicate the power, start time, and end time of user line i.

As shown in FIG. 7, the input command section 7 includes a microprocessor 71, a memory 72, and a detection information input section 7.
3. Input request information input section 74, response information output section 7
5, recognition information output section 76. The microprocessor 71 operates according to the flowchart shown in FIGS. 8a, b, and c. That is, first, in block 701, the usage status Ri,
Wi is reset (Ri=0 is a state in which recognition has not started, Wi=0 is a state in which no voice has been detected before starting recognition), and then in block 702 the voice detecting section 3 detects the detection information (user line i, Input the start time t _s and end time _te ). If the detected information is start end information, block 703 transfers the recognition unit input line ji (the user line is switched to the changeover switch 4) to the recognition unit 5.
start time t _s of the recognition unit input line connected by
or before the time when the user line i is connected to the recognition unit 5, and stores that state (Wi=1). If the detected information is termination information, the termination time _te of the recognition unit input line ji is notified to the recognition unit 5 in FIG. 8b, or if Wi=1, the response information (use user line i, voice input prompt sound as response content)
is output to instruct the recognition unit 5 to start recognizing the recognition unit input line ji. Furthermore, as shown in FIG. An instruction is given to start recognition, and the state in which recognition has started (Ri=1) is stored on the recognition input line ji.
By repeating the processes of blocks 702 and 703 and FIGS. 8b and 8c described above, the processes for m user lines are executed.

Control unit 8 is a US patent issued in March 1983.
USP4385359 “Multiple-Channel Voice Input/Output System”
The system can be configured as shown in FIG. 2 of the specification of ``System'', and operates according to the flowchart shown in FIG. 4 of the same specification.

The embodiment described above is used in common with a telephone line and an upper system in which an application program runs.
As shown in Figure 2, when a user calls the system by telephone, the telephone line control unit (NCU) 10i
detects the call and notifies the higher level system 200. After notification from the host system 200, voice input commands and voice output commands are output to the control unit 8 for the user line 1i to provide voice input/output services.

Although the present invention has been described above based on the embodiments, it is clear that a configuration in which m or n voice detection units, recognition units, and response units for one line are arranged can also be adopted. In addition, in this example, both the input command section and the control section are configured using microprocessors, so if the CPU has sufficient processing capacity, the CPU and memory can be shared and the number of CPUs and memory can be reduced to one. It is possible.

<Effects of the Invention> According to the present invention, a changeover switch is provided before the recognition processing section, and the recognition processing section is used in time division multiplexing.
By notifying the user of the timing to use the voice input by outputting a voice input promotion sound,
The number of lines simultaneously processed by the recognition processing section can be set to n, which is smaller than the number of user lines m, and the device can be made smaller.

In addition, a voice detection unit is installed in front of the switch to constantly monitor the voice input of the user's line, and if the user cannot wait to hear the voice input prompting sound and utters voice, the system is controlled so that voice input is performed again. By providing an input command section to do this, even if the user impatiently utters a voice as a signal, there will be no misrecognition or input not being accepted, and it is possible to provide the user with voice input that is easy to use.

[Brief explanation of the drawing]

Fig. 1 is a diagram showing the operation of the present invention, Fig. 2 is a block diagram showing an embodiment of the invention, Fig. 3 is a block diagram of a speech recognition device, and Fig. 4 is a diagram showing an embodiment of the detection section of the invention. 5a and 5b are flowcharts showing the operation of the detection section of the present invention, FIG. 6 is a diagram showing an example of voice detection operation, and FIG. 7 is a diagram showing an embodiment of the input command section of the present invention. The block diagram shown in Fig. 8a,
b and c are flowcharts showing the operation of the input command section of the present invention. In the figure, 11, 12, ..., 1m are user lines, 21, 22, ..., 2n are recognition section input lines, and 3
is a voice detection section, 4 is a changeover switch, 5 is a recognition section, 6
1 is a response section, 7 is an input command section, 8 is a control section, 31 is a preprocessing section, 32 is a microprocessor section, 10
1, 102, 10i, and 10m are NCUs, and 200 is an upper system.

Claims (1)

[Claims] 1 m (positive integer) user lines, a response unit that outputs audio to the m user lines, and the m
a detection unit that detects voice input from the number of user lines; a changeover switch that connects the output of the m voice patterns of the detection unit to n (positive integer) recognition unit input lines smaller than m; a recognition unit that recognizes the voices of n input lines that receive the output of the changeover switch; and a recognition unit that connects a user line that is in a talking state to the idle line of the recognition unit only when the presence of an idle line to the recognition unit is detected. a control unit that generates a control signal to generate a voice input request and outputs a voice input request; and a voice output that prompts the user line to speak when voice is not detected by the detection unit before the time when the voice input request is received. An input that commands the recognition unit to start by outputting a voice prompting the user to speak to the user line at the time when the end of the voice is detected when voice is being detected. A multi-line audio input/output device characterized by having a command section.