JPH0310560Y2 - - Google Patents

Description

[Detailed explanation of the idea] [Technical field of invention] The present invention relates to a speech recognition device.

[Prior art and its problems] Conventionally, in speech recognition devices for specific speakers,
Three modes: registration mode, test mode,
Recognition mode can now be set. The above test mode is a mode to check whether the registered pattern is appropriate.If the same registered word is input again and the distance to the pattern is less than the threshold,
The registered pattern is left as it is, and when it exceeds the threshold value, it is rewritten to the re-input audio pattern. In this case, voice input and pattern rewriting are repeated until the distance becomes equal to or less than the threshold value. The pattern is rewritten as described above, but in the conventional method, the more words are registered, the more time and effort is required during registration and test mode, which is extremely troublesome. Furthermore, since human voices change over time, it is necessary to re-register them, but it is extremely troublesome to re-register all words each time. Furthermore, if the registration is unclear or if a pattern that is easily mistaken is registered, erroneous recognition is likely to occur.

[Purpose of the invention] The present invention has been made in view of the above points, and provides a speech recognition device that can reduce errors during registration, improve the recognition rate, and reduce the trouble of re-registering speech patterns. The purpose is to provide

[Key points of the invention] This invention is a speech recognition device configured to prohibit the registration of words that are easily misrecognized due to their short distance, and to rewrite standard patterns in response to changes in speech over time. be.

[Embodiment of the invention] An embodiment of the invention will be described below with reference to the drawings. In FIG. 1, reference numeral 11 is a microphone, and its audio input is sent to a sound processing section 12. The acoustic processing section 12 converts the input audio signal into a digital signal and outputs it to the feature extraction section 13. A start/end detection section 14 is connected to the feature extraction section 13, which detects the start/end of the input audio from the level of the input signal and outputs the detection signal to the control section 15. The feature extractor 13 extracts the features of the input voice and temporarily stores them in the buffer 16 as a voice pattern. The audio pattern stored in the buffer 16 is selected by the selector 17 and sent to the standard pattern memory 18 or the distance calculation section 19.
The selector 17 operates according to a command from the control section 15. Further, the standard pattern memory 18 and the distance calculation section 19 are controlled by the control section 15. Standard patterns for words to be recognized are written in the standard pattern memory 18, and the stored patterns are sequentially read out to the distance calculation section 19 in response to instructions from the control section 15. This distance calculation section 19
compares the input voice pattern with the standard pattern, calculates the distance therebetween, and outputs the distance to the minimum value selection section 20. This minimum value selection section 20 is controlled by the signal section 15 and selects the minimum value from among the calculation results of the distance calculation section 19. The minimum value selected by the minimum value selection section 20 is sent to the control section 15 and also sent to the first threshold value determination circuit 22 or the second threshold value determination circuit 23 via the selector 21. The selector 21 performs a selection operation in response to a command from the control section 15. The first threshold value determination circuit 22 and the second threshold value determination circuit 23 are preset with threshold values TH1 and TH2 for the minimum distance between the audio pattern and the standard pattern. The above threshold values TH1 and TH2 are TH1<
By increasing TH2 in the TH2 relationship,
The allowable range for rewriting standard patterns when recognizing input speech has been expanded. Then, the determination results of the first threshold determination circuit 22 and the second threshold determination circuit 23 are sent to the control section 15. A key input section 24 and a display section 25 are connected to the control section 15, and the word corresponding to the result selected by the minimum value selection section 20 is displayed on the display section 25. The above key input section 24
When a correction operation is performed on the result displayed on the display section 25, a correction signal is sent to the control section 15 via the signal line a.

Next, the operation of the above embodiment will be explained with reference to the flowcharts of FIGS. 2 and 3. In the registration mode, when speech is input word by word from the microphone 11, it is converted into a digital signal by the sound processing section 12 and sent to the feature extraction section 13. Thereby, the feature extraction unit 13 extracts the features of the input voice and outputs them to the buffer 16. On the other hand, the start end end detection section 1
4 detects the start end of the audio signal from the signal input level in the feature extraction section 13 and outputs it to the control section 15 . The control section 15 starts reading the audio pattern output from the feature extraction section 13 into the buffer 16 in response to the start edge detection signal. When the end detection signal is output from the start/end detection section 14, the control section 15 transfers the audio pattern held in the buffer 16 to the distance calculation section 19 via the selector 17, and performs the steps shown in FIG. As shown in A1, the distance from each standard pattern stored in the standard pattern memory 18 is calculated. This distance calculation section 19
The calculation result is sent to the minimum value selection section 20, and the minimum distance is selected as shown in step A2. The selection result of the minimum value selection section 20 is sent to the control section 15, and the corresponding word is displayed on the display section 25. Further, the selection result of the minimum value selection section 20 is sent to a first threshold value determination circuit 22 via a selector 21.
As shown in step A3, it is determined whether the minimum distance is smaller than a threshold value TH1. If the minimum distance value is smaller than the threshold, that is, if the input voice is similar to the registered word in the standard pattern memory 18, the process proceeds to step A4, and the display section 25 instructs to re-input the voice. If it is determined in step A3 that the minimum distance is greater than or equal to the threshold TH1, that is, if it is determined that there are no other similar words, the process proceeds to step A5 and the selector 17 is stored in the standard pattern memory. 18 side, and writes the pattern of the input voice into the standard pattern memory 18 as a standard pattern. This completes the registration of standard patterns for one word. Thereafter, pattern registration for other words is performed in the same manner.

Next, a description will be given of the operation when recognizing the input voice after the above-mentioned pattern registration. In the recognition mode, when speech for one word is input, as shown in step B1 of FIG. 3, the distance calculating section 19 calculates the distance between the input speech pattern and the standard pattern, as in the above case. Next, the process proceeds to step B2, in which the minimum distance is determined in the minimum value selection section 20, and the minimum distance is determined via the selector 21 in the second threshold determination circuit 23.
sent to. In this second threshold determination circuit 23, as shown in step B3, the minimum distance is set to the threshold TH2.
It is determined whether the minimum distance is less than or equal to the threshold value TH2, and if the minimum distance is greater than the threshold value TH2, the recognition result is rejected as shown in step B4. If it is determined in step B3 that the minimum distance is less than or equal to the threshold TH2, the process proceeds to step B5, and the word corresponding to the minimum distance value is displayed on the display section 25 as a recognition result. If the recognition result is different from the user's intended result, the user issues a correction instruction from the key input unit 24. When a correction signal is input from the key input section 24, the control section 15 instructs re-input on the display section 25 as shown in step B7. Further, in step B6, if no correction signal is input, the pattern of the input voice held in the buffer 16 is transferred to the standard pattern memory 18 via the selector 17 as a standard pattern.
Register. That is, in the recognition mode, if the recognition result is correct, the pattern of the input voice is re-registered in the standard pattern memory 18 as a standard pattern. This standard pattern memory 18
Pattern registration is performed each time recognition is performed.

[Effects of the invention] As described above, according to the invention, in the registration mode, words that are easily misrecognized due to their small distance from other words are not registered in the standard pattern memory 18. It is possible to reduce errors during registration and improve recognition rate. In addition, in recognition mode, if the recognition result is correct, the pattern of the input voice is re-registered in the standard pattern memory 18, and the range of recognition as correct is expanded, so the frequency of rewriting the standard pattern is reduced. This makes it possible to prevent erroneous recognition due to changes in voice over time, and to reduce the effort required to re-register standard patterns.

[Brief explanation of drawings]

The drawings show one embodiment of the present invention; FIG. 1 is a block diagram showing the circuit configuration, FIG. 2 is a flowchart showing the operation in registration mode, and FIG. 3 is a flowchart showing the operation in recognition mode. It's a chat. DESCRIPTION OF SYMBOLS 11...Microphone, 12...Acoustic processing section, 13...Feature extraction section, 14...Starting end end detection section, 15...Control section, 16...Buffer, 17...
...Selector, 18...Standard pattern memory, 19...
... Distance calculation section, 20 ... Minimum value selection section, 21 ...
Selector, 22...First threshold value determination circuit, 23...
Second threshold value determination circuit, 24...key input section, 25...
...Display section.

Claims (1)

[Scope of Claim for Utility Model Registration] Distance calculation means for calculating the distance between the input voice and each standard pattern registered in the standard pattern memory and taking the minimum value among the calculation results; The minimum value is the first
a first determination means for determining whether or not the minimum value obtained by the distance calculation means is equal to or greater than a second threshold set larger than the first threshold; a second determining means for determining the voice pattern; and a second determining means for determining the voice pattern of the input voice as a new standard pattern when the minimum value regarding the input voice at the time of registering the voice pattern is determined by the first determining means to be equal to or greater than the first threshold value. a control means for registering the standard pattern in the standard pattern memory and instructing re-input of the voice when it is determined that the voice is smaller than the first threshold; a recognition means that selects a standard pattern corresponding to the minimum value as a recognition result when it is determined that the minimum value is less than the threshold of 2, and outputs a word related to the standard pattern; and when the word output by this recognition means is appropriate; 1. A speech recognition device comprising: standard pattern rewriting means for rewriting a standard pattern selected by the user into an input speech pattern.