JPH0459639B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a recognition device that is effective when applied to, for example, a speech recognition device that includes a plurality of specified frequency response sensors and a plurality of processing devices.

Conventionally, voice recognition systems use a single sensor that covers a wide range of frequencies to detect the target sound, a frequency analyzer separates each frequency, and a single computer processes this using software to create a standard frequency pattern. It recognized speech by comparing it to patterns.

This system has the disadvantage that it is difficult to detect and correct failures in the case of sensor failures or frequency analyzer failures, and that voice recognition becomes completely impossible in the case of computer failures.

The purpose of the present invention is to use a simple algorithm to enable multiple microcomputers to quickly and reliably recognize objects without failing to recognize them or erroneously recognizing them, even if a partial failure occurs. The goal is to provide equipment.

In order to achieve this object, the present invention includes a large number of detection means for detecting partial information of a target, which is divided into parts such that some parts overlap, and , a plurality of processing means each processing a combination of signals from a plurality of detection means whose detection outputs are partial detection information that partially overlap with each other; and a recognition result regarding the entire object based on the processing results of these processing means. The present invention is characterized in that the processing means is configured to process the detection information from the detection means using changeable parameters.

Further, in the present invention, a plurality of frequency sensors each having a different center frequency of the sensing frequency band are provided, and each sensor is configured to detect only a signal in a predetermined frequency band and transmit the signal to a corresponding processing device. There are characteristics. Each processing device receives this signal from the directly connected sensor and the adjacent sensor, performs processing based only on these signals, and determines the frequency. In this case, adjacent sensors are configured to detect frequencies that partially overlap, so even if some of them fail, they will not become unrecognizable.

Hereinafter, the present invention will be explained in detail based on examples. FIG. 1 is an overall configuration diagram of a speech recognition device according to the present invention. Frequency sensors 1 to 7, each having a different sensing frequency band, are connected to interfaces 10 to 7.
0, and are connected to recognition processing devices 100 to 700 via bus transmission lines 10a to 70c. Furthermore, the recognition processing device uses the bus transmission line 1000 to
It is connected to the audio output device 2000.

For example, the recognition processing device 200 includes frequency sensors 10, 20, 30 and common bus transmission paths 10c, 2.
They are connected to each other via sensors 0b and 30a, and voice recognition is performed based on the outputs of the three sensors 1, 2, and 3.

FIG. 2 shows the internal structure of the recognition processing device 200.
The processing device 202 includes bus transmission lines 10c, 20b,
30a is sequentially fetched by opening and closing the interface 201 at regular intervals based on the timer 203, and stored in the corresponding reception buffers 204, 205, and 206, respectively. The processing device 202 puts the processing results for these inputs into the transmission buffer 207 and further sends them to the interface 20 under the timer 203.
8 to the audio output device 2000.

FIG. 3 shows the internal structure of the audio output device 2000, in which the processing device 2002 transfers data sent via the bus transmission path 1000 to the interface 2.
001 is taken in by opening and closing, and stored in the reception buffer 2004. The processing device 2002 is
A frequency is calculated from a series of received data and transmitted to the sound generation device 2005 to generate sound. Alternatively, the recognized frequency is stored in the storage device 2006.

Next, we will explain the operation from sound recognition to output.

For example, frequency sensors are 1, 2, 3, 4,
5, 6, 7 are respectively 500~550Hz, 525~575Hz,
550~600Hz, 575~625Hz, 600~650Hz, 625~675
Hz, and the response to frequencies of 650 to 700 Hz, and the input/output characteristics of each are shown in FIG. Suppose now that a 600Hz sound comes into this device. At this time, since there is an input/output characteristic as shown in Fig. 4, sensors 3, 4, and 5 each have v=
Recognition processing device 30 corresponding to outputs 1, 2, 1
Transfer to 0,400,500. Other sensors are not responding and output is 0 level. On the other hand, the recognition processing device 200 receives signals from the sensors 1, 2, and 3 through the bus transmission lines 10c, 20b, and 30a, and processes the input data using the following algorithm.

The output P _j of each recognition processing device j is three input data
Calculate as follows based on v _i-1 , v _i , and v _i+1 .

P _j =k _i v _i-1 +k ₂ v _i +k ₃ v _i+1 where k ₁ , k ₂ , and k ₃ are constants, and can be appropriately determined according to the output characteristics of the sensor. Generally, k ₂ =1, k ₁ =k ₃ =-0.5, and the output value of that portion is suppressed by the output values of the sensors on both sides. This is to avoid overlooking parts where the input changes.

With this algorithm, the recognition processing device 100
~700 are respectively 0, -0.5, 0, 1, 0-
Outputs 0.5, 0. FIG. 5 shows the correspondence between this output result and each sensor and recognition processing device. This result is displayed on the audio output device 2000.
sent to. If the processing device 2002 is a convex portion and the front and back of the convex portion are surrounded by 0 values, the processing device 2002 regards the central portion of the convex portion as the input frequency, and the audio generating device 200
5 or to the memory 2006. Here, data is transferred from the recognition processing device to the audio output device only from the recognition processing device that has a non-zero output value, but the use of the bus transmission line 1000 is permitted to avoid collisions during data transfer. Audio output device 2
000 will do it.

The above explanation is for a normal case, but below we will explain a case when some of the sensors and recognition processing devices are out of order.

(1) Failure of frequency sensor When sensor 4 fails in FIG. 5, recognition processing devices 300, 400, and 500 no longer receive sensor output from sensor 4, and perform processing using only two inputs. As a result, as shown in Figure 6, the outputs of the recognition processing device are 0, -0.5, 1, -1, 1, -0.5, 0.
becomes. The processing device 2002 in the audio output device detects two convex parts, but since they are not surrounded by 0, it does not judge that they are two frequencies, but instead uses the central frequency surrounded by 0 (sensor, band 4...
600Hz) is determined as the input frequency. (if,
If there is a one-to-one correspondence between the sensor and the processing device, the band of the sensor 4 cannot be recognized. ) In this way, even in the case of a sensor failure, the correct frequency band can be recognized by this method.

(2) Failure of recognition processing device Now suppose that the device 400 has failed. At this time, the inputs coming to the audio output device 2000 are 0, -0.5, 0, 0, 0, as shown in FIG.
-0.5. Although the sharpness of the convex portion is lost, the input frequency can be recognized as the central portion of the convex portion surrounded by zeros. As described above, even if the sensor or recognition processing device fails, the input will not be ignored or recognized incorrectly.

Finally, I will explain the optional sound selection function.
Normally, the recognition device described above receives input from all sensors, detects some protrusions, and recognizes speech. However, when it is desired to recognize only a certain voice or to identify sounds in a certain frequency range, the processing unit 2002 in the audio output device in FIG. In response, a command is issued not to send data. The recognition processing device that receives the command continues to maintain an output state of 0 level regardless of the output value from the sensor. As a result, the audio output device 2000 can receive only the output value of the frequency portion that it wants to receive, and can arbitrarily select the sound.

As explained above, according to the present invention, frequencies can be recognized by arranging a large number of sensors with slightly different frequency detection ranges and connecting each to a microcomputer. can be avoided, and high reliability of the system can be achieved. Furthermore, it is also possible to arbitrarily select only the necessary voices or bands.

[Brief explanation of drawings]

Fig. 1 is an overall configuration diagram of the speech recognition device according to the present invention, Fig. 2 is a detailed block diagram of the recognition processing device, Fig. 3 is a detailed block diagram of the speech output device, and Fig. 4 is the input/output of one frequency sensor. Characteristic diagram, Figure 5
Figures 6 and 7 are diagrams showing the output level of each processing unit when a 600Hz sound is generated, and Figures 6 and 7 are diagrams showing the output level of each processing unit when the frequency sensor and recognition processing unit fail, respectively. . 1~7...Frequency sensor, 100~700...
...Recognition processing device.

Claims (1)

[Scope of Claims] 1. A large number of detection means each detecting information on a target that is divided so that a portion of the information overlaps, and among the detection means, a portion of which overlaps each other. A plurality of processing means each processing a combination of signals from a plurality of detection means each having partial detection information as a detection output, and an output means outputting a recognition result regarding the entire object based on the processing result of the processing means. A recognition device, characterized in that the processing means is configured to process detection information from the detection means using changeable parameters. 2. Frequency detection means provided for each partial band to detect the band frequencies of a plurality of partial bands formed by dividing the frequency band of the input audio signal so that some parts overlap each other, and a frequency detection means provided for each partial band, and a frequency detection means provided for each partial band, and a frequency detection means provided for each partial band to detect the band frequencies of a plurality of partial bands formed by dividing the frequency band of the input audio signal so that some parts overlap each other. processing means for receiving the output signals of the plurality of detection means for detecting the band frequencies of and recognizing the frequencies in the corresponding partial bands; and means for recognizing the input audio signal based on the processing results of the processing means. A recognition device characterized by comprising: