Research And Realization On Remote Control System Based On Speech Recognition

With the realization of instructing machines to complete a series of controls based on human commands, Human computer interaction has become new trend on technology development. Researches on speech recognition evolve rapidly both at home and abroad and even some tech-companies join into this research team as well.Along with popularity of Internet, speech recognition can control machines remotely by network. Application with this technique has already been used in industrial production, intelligent agricultural management system and home intelligent management. In this context, this project researches isolate-word speech recognition technique combined with remote control based on embedded systems.This essay first describes the five basic components of the framework of isolate-word speech recognition system, which includes pretreatment process of voice signals, endpoint detection algorithm, feature extraction, model training and recognition. For the endpoint detection algorithm, analysis are made beforehand on the unsatisfying result when applies time domain or frequency domain parameters independently so that this research decides to combine both parameters to determine the start and end point of signals. In the part of feature extraction, MFCC feature parameters are described specifically. At last in the isolate-word speech recognition algorithm, this essay describes dynamic time warping as core and comes up with improvements to the problem of its huge computation amount.In hardware design, this project designs handheld recognition end based on DSP chips and remote control end based on STC80C52. Also, application of GSM modules realizes communication and then builds an integrated remote-control speech recognition system.?In software design, with design of each part, this project gives an integrated process of voice signal collection, pretreatment, endpoint detection and recognition.As a last part, the project makes test by sending commands to GSM modules,detecting effects at endpoints as well as the rate of recognition. Therefore evidences are found to ensure the effectiveness and feasibility of the system.