Method for frequency selection circuit to trigger multilevel multiword voice wakeup control lamp

Abstract

The invention relates to the application field of the technical field of voice recognition control and provides a method for a frequency selection circuit to trigger a multilevel multiword voice wakeup control lamp. A system is composed of a universal MCU, an independent 16-bit ADC chip, a high sensitivity and high signal to noise ratio microphone, an ordinary microphone, a signal processing circuit of the ordinary microphone, a dip switch and a peripheral circuit. Software is a multilevel multiword voice wakeup engine. According to the method, through adoption of 11 steps, functions of turning on, turning off, dimming and toning the lamp can be realized only by speaking two sentences: a first level wakeup word and a second level wakeup word, so a habit of a user is cultivated effectively,and two hands of the user are free actually. According to the frequency selection circuit provided by the invention, the system is enabled to work in a low power mode at ordinary times and when the user speaks, the multilevel multiword voice wakeup engine is triggered, so the effect of saving energy, reducing spurious triggering and improving accuracy is achieved.

Description

technical field [0001] The invention relates to the technical fields of voice recognition control and lamp control, and provides a method for controlling a lamp by triggering a multi-stage multi-word voice wake-up by a frequency selection circuit. Background technique [0002] Speech recognition is currently showing an oligopoly pattern both domestically and internationally. In the international market, Nuance's market share is far ahead, accounting for 62% of the market share. In the domestic market, HKUST Xunfei has obvious advantages, occupying 54% of the market share and ranking first. [0003] At present, although the speech recognition technology at home and abroad has made great progress, there are still many limitations in the application. Speech recognition is generally divided into cloud-based non-specific human voice large-scale continuous speech recognition, non-network voice wake-up, and commands that can not be connected to the Internet. word recognition. ...

AI Technical Summary

Problems solved by technology

[0004] (1) Large-scale continuous speech recognition of non-specific human voices in the cloud, with high recognition accuracy, can add semantic understanding, NLP, and TTS functions, but requires devices to be connected to the Internet, slow response time, high cost, and poor user experience

[0005] (2) Voice wake-up means that the wake-up words are always detected in real time, and the user experience is good. However, when there are multiple wake-up words, the probability of false wake-up will increase. At the same time, the wake-up words need special customization on different MCUs and cannot be used universally.

[0006] (3) Command word recognition is to make it work through some kind of trigger such as key press or voice wake-up, record the sound for 3-5 seconds and then recognize it. This method is relatively mature, but the CPU and RAM occupied during operation will increase significantly. At the same time, command words need special customization on different MCUs, and cannot be used universally

[0007] (4) The existing voice recognition modules that are not connected to the Internet are basically operated by ARM11 and above CPUs, which are large in size, high in cost, high in power consumption, and low in recognition rate. Linux, the stability of the system is much worse than that of ordinary MCU, so it is not suitable for use on lamps

[0008] (5) The existing non-networked speech recognition modules all need to do Fourier operations, which consume a lot of power, and when there is no one, they are always doing recognition, that is, it is not energy-saving, and it is not necessary

Images