KR20000001476U - Design of door lock device by speaker recognition - Google Patents

Abstract

Since the voice is different for each individual, the voice of each individual is encrypted and the door lock device is controlled by using the voice instead of the key without using the door key. To this end, speaker recognition technology is needed to determine whether or not a specific sentence voice input through a microphone is a standardizer.

LPC cepstrum is used as a feature to obtain the personal features included in the speech to be used in speaker recognition technology, and vector quantization for the grouping of extracted personal features is used as a method of collecting the representative features of each standardizer. By using speaker recognition, DTW is used to first select whether a voice of an arbitrary speaker corresponds to a standardizer, and is encoded into a codebook in which individual characteristics are collectively subjected to vector quantization from the voice of the standardizer selected by the DTW in advance. Secondly, discrete HMM and Viterbi algorithms are performed to confirm that they match those of the initially selected standardizer.

If both of these processes are met, it confirms that the specific sentence voice is the standardizer and controls the door lock device.

This is a device that controls the door lock by using the security function as the voice pattern of the human voice and the promised body sentence, which is more convenient in terms of convenience than the security access device using the existing RF ID card, etc. It is mainly used for doors that require security because it is advantageous in installation cost than security access device and iris access device.

Description

Design of door lock device by speaker recognition

Since the voice is different for each individual, the voice of each individual is encrypted and the door lock device is controlled by using the voice instead of the key without using the door key. For this purpose, the speaker recognizes whether or not the specific sentence voice input through the microphone is the standardizer. As a result, it is determined whether or not the specific sentence voice inputted by the microphone is the standardizer and controls the door lock device.

By using the voice of the standardizer as a cipher, the speaker recognition of a specific sentence voice is performed to control the door lock device without the door key. The key techniques used to perform speaker recognition are LPC cepstrum [1], vector quantization [2], DTW [3], and HMM and Viterbi algorithm [1].

(1) conventional speaker recognition technology

The conventional speaker recognition structure is shown in Figure 1. That is, the feature is extracted from the speech waveform and compared with the standard pattern of the speaker registered in advance. The recognition result is determined by distance or similarity like voice recognition.

If the distance between the input voice pattern and the standard pattern is smaller than the threshold value, it is recognized as a registered speaker. Otherwise, it is not recognized as a registered speaker.

The following figure 2 shows the result of speaker recognition with specific sentences spoken by 100 men and women speakers. Where the horizontal axis is the number of speakers and the vertical axis is the false recognition rate.

Characteristic characteristics of voice-unique personality include spectral parameters LPC cepstrum Dynamic features, Was used.

here, Is Symmetrical window of length or one. The effective formula for the distance measurement of this feature is as follows.

Also, in this equation silver Weighting factor inversely proportional to the variance of and Are the characteristics of the normalizer and the input unknown speaker respectively.

(2) Speaker recognition technology and door lock control

The flowchart of the speaker recognition system to be proposed is shown in FIG. First, when the speech waveform is input, the spectral features are extracted by LPC analysis, and the time base is normalized using the DTW. During training, the codebook obtained by vector quantization is quantized into a codebook of registered speakers. At this time, the candidate quantizer is selected by quantizing the codebook with the minimum quantization distance. A Viterbi algorithm is performed to compare the quantized code sequence with the candidate speaker's codebook with the HMM of each speaker.

The speaker of the HMM whose maximum probability is obtained in this algorithm is the speaker recognized by the system. Proposed speaker recognition system reconfirms time-series feature vectors by adding HMM to 97% of existing systems obtained from conventional speaker recognition, so it performs more than 97% speaker recognition. Therefore, the speaker recognition result is connected with the door lock to control the door lock.

LPC cepstrum is used as a characteristic by obtaining the speaker's personal characteristics, and vector quantization is used as a method of collecting representative characteristics of each standardizer. As shown in the flowchart of FIG. 2, the DTW is primarily used to select whether the voice of the randomizer corresponds to the standardizer, and is encoded and coded into a codebook obtained through vector quantization in advance from the voice of the first selected standardizer. Perform an HMM to secondary verify that it matches the first chosen standardizer. If both of these processes are met, it confirms that the specific sentence voice is the standardizer and controls the door lock device.

1 is an overall block diagram for controlling the door lock device by determining whether or not a specific sentence voice of an arbitrary speaker spoken through a microphone is a standardizer.

2 is a flowchart for speaker recognition of an arbitrary voice spoken by an arbitrary speaker.

Speaker recognition technology is used to control the door of the device by the voice of the person. As shown in the entire block diagram of FIG. 1, the visitor's monitoring device may be divided into a microphone, a computer, and a door lock device. When a microphone speaks a specific sentence that a person is supposed to use for speaker recognition, it is transmitted through a computer-connected electrical cable. In the computer, the input voice is compared with the characteristic pattern of the pre-recorded voice of the standardized person who is allowed to enter through the software configured according to the flowchart for speaker recognition of a specific sentence voice, and if the pattern matches, the electric cable from the computer An electrical signal is sent through the furnace to the door lock device to open the door lock device.

It is a device that performs door lock function by security function with the voice pattern of human voice and the promised body sentence, which is more convenient in terms of convenience than security access device using existing RF ID card, security access device using fingerprint recognition, It is mainly used for doors requiring security because it is advantageous in installation cost rather than iris access device.

Claims (2)

In the present invention, when the voice of a predetermined sentence is uttered and input into a computer, speaker recognition is performed to determine whether the voice is a predetermined standardizer or not, thereby controlling the opening and closing of the door lock device. The speaker recognition according to claim 1, wherein the speaker recognition is made by software configured in accordance with a flowchart for speaker recognition of a specific sentence voice of FIG. * references [1] L. R. Rabiner and B. Gold, Fundamentals of Speech Recognition, Prentice Hall International, Inc., 1993 [2] Y. Linde, A. Buzo, R. M. Gray, "An Algorithm of Vector Quantization Design," IEEE Trans. Communication, Vol. COM-28, pp. 83-108, 1980 [3] H. Sakoe, S. Chiba, "Dynamic Programming Algorithm Optimization for Spoken Word Recognition," IEEE Trns. Acoust. Speech, Signal Processing, Vol, ASSP-26, No. 1, pp. 43-49, 1978