JPH01216398A - Speech recognizing system - Google Patents

Abstract

PURPOSE:To input a voice by one set of speech recognizing device against plural voice input persons by inputting the voices from each channel by a time division by determining the priority when the voices are inputted simultaneously from plural channels. CONSTITUTION:An interface circuit takes OR of voices inputted from voice input devices 11-1n of each channel and inputs it to a speech recognizing device 2, and simultaneously, voice output period detecting devices 51-5n detect a voice output period of the voice input devices 11-1n. Subsequently, a voice input channel deciding means specifies an input voice channel by a detecting signal, and also, when voices are inputted simultaneously from plural channels, a voice input control means outputs a message for instructing to a voice input person so that the voice is inputted to the voice recognizing device 2 in order from a prescribed channel, to response devices concerned 41-4n, brings the voices inputted simultaneously to time division and inputs them to the speech recognizing device 2. In such a way, a voice input can be executed to a computer 3, etc., by one set of speech recognizing device 2 against plural voice input persons.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial Application Field) The present invention relates to a speech recognition method in which a plurality of speech inputs are recognized and processed by a speaker-independent word recognition device.

(Prior Art) Conventionally, systems have become commonplace in which events that occur during a work process in which multiple workers perform similar tasks at industrial sites are input into a computer or the like by voice and the information is processed. In such a system, the contents input by voice are represented by simple words in advance, and the words input by each worker are recognized by a voice recognition device installed for each worker. A voice recognition method is adopted in which the recognition result is sent to the host computer. The host computer performs information processing based on the sent recognition results, sends answers to the worker as appropriate, and proceeds with a series of tasks. The means by which the host computer sends the answer to the worker include a method in which the answer is sent to the worker by voice through a response device, a method in which the answer is displayed on a display device such as a CRT, and the answer is presented to the worker.

FIG. 6 is a block diagram showing an example of a conventional voice input response system. Voices generated by each worker are transmitted through microphones 11 to 1. from there to the corresponding speech recognition devices 21 to 2o. The results of voice recognition by each voice recognition device @21 to 2o are sent to the host computer 3, and the host computer 3 identifies from which voice recognition device the input voice recognition result was input. Then, processing corresponding to the voice recognition result is performed and the result is displayed on the corresponding CRT as an answer. Therefore, when a voice is input from the microphone 1, the answer is displayed on the CRT 41, and the worker proceeds with the work by looking at this answer.

However, depending on the content of the work handled by the above-mentioned conventional systems, the frequency of occurrence of events is low, and even if multiple voice recognition devices are installed, the probability that voice input will occur at the same time is extremely low, and two or more voice recognition devices In some cases, the two rarely operated at the same time. In such a case, the operating rate of the speech recognition device is extremely low, and it is extremely uneconomical to install a plurality of expensive speech recognition devices.

(Problems to be Solved by the Invention) In the conventional speech recognition method described above, a plurality of speech recognition devices are required to match the number of speech inputters even when the frequency of speech input is low. For this reason, the operation rate of each voice recognition device is poor, resulting in a disadvantage that it is extremely uneconomical. Therefore, the present invention aims to eliminate the above-mentioned drawbacks, and aims to provide a voice recognition method that allows multiple voice input users to input voice to a computer, etc. using one voice recognition device. .

[Structure of the Invention] (Means for Solving the Problems) The present invention has a plurality of channels each consisting of a voice input device and a response device, and a voice recognition device recognizes the voice input from each channel. In a system that has a computer that sends a response based on the recognition result to the response device of the input voice channel, there is one voice recognition device, and the logical sum of voices input from the voice input devices of each channel to this voice recognition device is used. an interface circuit that takes and inputs the
A plurality of audio output period detection devices correspond to each audio input device on a one-to-one basis and detect the audio output period of the corresponding audio input device, and an input audio channel is detected by a detection signal input from the audio output period detection device. an audio input channel determining means for determining whether or not audio is input from multiple channels at the same time; A configuration comprising a voice input control means that outputs a message instructing a voice input person to input to the corresponding response device, and controls the simultaneously input voices to be time-divided and input to the voice recognition device. have.

(Operation) In the speech recognition system of the present invention, the interface circuit calculates the logical sum of the voices input from the voice input devices of each channel and inputs the result to the voice recognition device. At the same time, the audio output period detecting device detects the audio output period of the corresponding audio input device and notifies the audio input channel determining means of the computer.

The audio input channel determining means specifies the input audio channel based on the detection signal input from the audio output period detection device, determines whether audio is input from multiple channels simultaneously, and sends the result to the audio input control means. Inform. The voice input control means outputs a message to the corresponding response device instructing the voice input person to input voice into the voice recognition device in order from a predetermined channel when voice is input from multiple channels at the same time. , the simultaneously input voices are time-divided and input to the voice recognition device.

(Embodiment) An embodiment of the present invention will be described below with reference to the drawings, in which the same parts as those of the conventional example are given the same reference numerals. FIG. 1 is a block diagram showing an embodiment of a voice input response system to which the voice recognition method of the present invention is applied.

Microphones 11 to 1o are installed corresponding to the number of voice inputters, and convert the voice of each voice inputter into an audio signal;
2 is a speech recognition device that recognizes the input speech signal (converts it into a corresponding code, etc.) and outputs the result to the host computer 3; 3 performs processing based on the speech recognition result input from the speech recognition device. The host computer displays the processing result on the corresponding CRT, 41 to 4o are CRTs that display the response of the host computer 3 to the voice input, and 51 to 5o detect the output period of the voice input from the corresponding microphone, A voice output period detection device 6 inputs the detection result to the host computer 3, and a voice recognition device 6 calculates the logical sum of the voice signals output from the microphones 11 to 1o.
This is an OR circuit that inputs to

Next, the operation of this embodiment will be explained. First, the operation when there is audio input from only one channel will be described. Assume that when an event occurs during a work process, a certain worker utters a word sound corresponding to this event into the microphone 11. The microphone 11 converts the collected voice into an audio signal and outputs this to the voice recognition device 2 via the OR circuit 6. The voice recognition device 2 performs recognition processing to recognize the input voice signal, converts it into a code that the pruritus meter can understand, and sends this code to the host computer 3 as a recognition result.
Output to. On the other hand, the audio signal output from the microphone 11 is input to the audio output period detection device 51,
Here, the audio output period is detected.

FIG. 2 is a block diagram showing a detailed example of the audio output period detection device. The audio signal 100 as shown in FIG.
The signal 20 as shown in FIG. 3(B) is half-wave rectified at
It becomes 0. By passing this signal 200 through the peak hold circuit 52, envelope information 300 of the audio waveform shown in FIG. 3(C) is extracted. This envelope information 300 is compared with a reference voltage V in a conveyor circuit 53. This conveyor circuit 53 outputs a logic "1P" signal 500 on the output side in a section where the input voltage is higher than the reference voltage ■.Here, if the reference voltage ■ is appropriately selected, it is possible to , noise sections with low levels can be distinguished, and signal sections where the input voltage is higher than the reference voltage V can be regarded as audio signal sections.Therefore, the conveyor circuit 5
The output 500 of No. 3 is as shown in FIG. 4.5 (B), where the voice section is "1" and the noise section is "0". The output 500 of the conveyor circuit 53 is ANDed with the clock 400 by an AND circuit 54, and then inputted to a retriggerable one-shot multivibrator 55. When a trigger pulse is input, the retriggerable one-shot multivibrator 55 generates a pulse 600 at the output for a time width determined by a time constant CR, and if the input trigger pulse is within the time width, the output is maintains the state of “1”.

Also, if the input trigger pulse stops, the output state becomes "Ote" after the above-mentioned time interval from the trigger pulse. Therefore, the input signal 500 and the clock 400 change as shown in FIG. 4.5 (B).
, (C), the output 60 of the conveyor circuit 53
0 is as shown in FIG. 4.5 (D). In other words, even if there is a valley in the middle of the input audio as shown in Figure 5 (A>) and this valley falls below the reference voltage ■, if the interval between these valleys is smaller than the time width, the effect of this valley will be on the output side. Therefore, even if the input audio fluctuates greatly, the audio output period detection devices 51 to 5o can stably detect the audio generation period.However, the period of the clock 400 is determined by the output pulse width. The output signal 600 of the audio output period detection device 51 is input to the host computer 3, and the host computer 3 identifies from which audio output period detection device the output signal 600 is input. By this, the host computer 3 knows that the voice has been input from the voice input channel, that is, the microphone 11 in this case.Thus, when the host computer 3 performs processing based on the recognition result of the voice recognition device 2, the host computer 3 receives the processing result. Send it as an answer to the CRT41 of the audio input channel,
display a response message such as the next work instruction.

In this way, when a certain channel is inputting audio, the host computer 3 can determine the channel number that is inputting audio based on the audio output period detection signal 600 output from the corresponding audio output period detection device. . Therefore, for CRT of other channels, audio 1iii1
A message A indicating that device 2 is in use is sent and displayed.

The message A may be, for example, "The voice recognition device is busy. Please wait for a while for voice input." Thereafter, when the host computer $1113 completes the input processing for the channel, it sends a message B as shown below, for example, to the CRTs of all channels. [The voice recognition device is ready. Voice input is possible. ” Next, if multiple channels perform audio input at the same time,
The output of the OR circuit 6 is a superimposition of multiple sounds,
This superimposed voice is input to the voice recognition device 2. Therefore, even if these voices are recognized by the voice recognition device M2, the vlh recognition results are not reliable.

However, in this case, multiple temporally overlapping audio output period detection images from the audio output period detection devices of multiple channels @
600 is input to the host computer 3. Therefore, the host computer 3 recognizes that there is voice input from multiple channels at the same time, discards the recognition result input from the voice recognition device 2, and prioritizes the multiple channels that input voice at the same time. decide.

Next, host computer 3 uses C, the channel with the highest priority.
Send the following message C to RT and have it displayed.

In other words, "Which voice should you say next?"

At the same time, the host computer 3 sends the same message A as described above to the CRTs of channels with other priority levels. In other words, "The voice recognition device is busy. Please wait for voice input for a while." When the voice input of the channel with the highest priority is completed, the host computer 3 sends message C to the channel with the second highest priority. , and processes to send message 8 to channels with other priorities. In the end, the host computer 3 repeats the above operation and performs audio input on a time-sharing basis for all channels to which audio has been input redundantly.
An effective recognition result is obtained from the speech recognition device 2, and speech input processing is sequentially performed. Note that the audio L1'lA device 2
Messages regarding the usage status of the CRT can be displayed in the corner of the CRT screen, and instructions such as work details can be displayed in a large size in the center of the screen.

According to this embodiment, when voices are input from multiple channels at the same time, priority is determined and the voices from each channel are input in a time-sharing manner, so there is only one voice recognition device 2. It is possible to configure a multi-channel audio input system at a very low cost.

Furthermore, in cases where voice inputs occur infrequently, voices are rarely input at the same time, and a practical voice input processing speed can be maintained.

[Effects of the Invention] As described above, according to the speech recognition method of the present invention, there is an effect that a plurality of speech input users can input speech into a computer or the like using one speech recognition device.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of a voice input response system to which the voice recognition method of the present invention is applied, and FIG.
FIG. 3 is a block diagram showing a detailed example of the audio output period detection device shown in FIG. Operation waveform diagram of the latter part of the audio output period detection device shown in Figure 2, No. 6
The figure is a block diagram showing an example of a conventional voice input response system. 11-1n...Microphone 2...Speech recognition device 3...Host computer 41-4o-CRT 51-5n...Speech output period detection device 6...OR circuit agent Patent attorney Noriyuki Chika Zhou Shanwang - Figure 2 jlI Figure 3 (C) Figure 4 (C) Figure 5 a)

Claims (1)

[Claims] A computer that has multiple channels consisting of a voice input device and a response device, has a voice recognition device recognize the voice input from each channel, and sends a response based on the recognition result to the response device of the input voice channel. In a system with a single voice recognition device, an interface circuit that calculates the logical sum of the voices input from the voice input devices of each channel and inputs the result to the voice recognition device, and a one-to-one interface circuit for each voice input device. a plurality of audio output period detection devices that correspond to each other and detect audio output periods of the corresponding audio input devices;
an audio input channel determining means for specifying an input audio channel based on a detection signal input from the audio output period detection device and determining whether or not audio is input from multiple channels simultaneously; is input, a message instructing the voice input person to input voice into the voice recognition device in order from a predetermined channel is output to the corresponding response device, and the simultaneously input voices are time-divided. A voice recognition system comprising voice input control means for controlling input to the voice recognition device.