Ultrafast Humidity Sensor Based On GOQD@SiO₂ Core-shell Structure And Its Application In Speech Recognition

Speech recognition in noisy environments has always been a difficult problem.At present,there are three main methods for speech recognition in noisy environments:speech enhancement,speech feature extraction,and adjustment of speech model parameters.These methods can resist noise to some extent,but they cannot solve the noise interference fundamentally.Therefore,a GOQD@Si O₂ core-shell structure ultra-fast humidity sensor was proposed in this paper.Considering the ultra-fast response speed and the noise insensitivity of the sensor,we designed an anti-noise speech recognition system using this humidity sensor as an acoustic-electric transducer.The humidity sensor,using the graphene oxide quantum dots（GOQD）as humidity sensitive material,is supported by staggered electrodes.And the strong cationic polymer polydiallyl dimethylammonium chloride（PDDA）was used as an assembly substrate.GOQD is assembled on the surface of silica microspheres by electrostatic self-assembly,and finally a GOQD@PDDA@Si O₂ core-shell structure is formed on the interleaved electrodes.The test results show that the sensor has a super-fast response speed of 3 milliseconds and a sensitivity of 33.47K/RH%.In addition,the micro-structure of the sensor and the characteristics of graphene oxide quantum dots were analyzed by us,we figured out the reason why the sensor has a super-fast response speed.Then we used AC complex impedance spectroscopy to reveal the humidity-sensitive mechanism of the sensor.Through the research and analysis of the response characteristics of the letters,we have proved the feasibility of ultra-fast humidity sensors in speech recognition.FPGA was used by the recognition system to control ADC to collect the signal of humidity change around the mouth during pronunciation,and identification system is transmitted to the computer via UART.First,in order to obtain the effective signal of the voice signal,the starting point and duration of the voice signal need to be processed by the system in the early stage to obtain.Second,in order to transform the time-domain voltage signal into the frequency-domain signal,the short-time Fourier transform of the signal was made by the system.Third,in order to project the signal into the low-dimensional space,reduce the dimension of the signal and obtain its feature vector,the principal component analysis of the signal is done by the system.Finally,in order to realize the real-time visual recognition of signals,the host computer based on MATLAB is used by the system to display the recognition results in real time.In the end,a test environment was set up to verify the anti-noise performance of the speech recognition system.During the experiments of recognizing some words in noisy and quiet environment,the accuracy of speech recognition system is more than 90%.A feasible solution for speech recognition in noisy environment is provided by the system.