Research On Continuous Speech Keyword Recognition Based On Time Domain And Frequency Domain

With the development of artificial intelligence in various fields of life,speech recognition has a broad application prospects as one of the important human-computer interaction.Continuous voice keyword recognition has become a research hotspot in recent years,with its fast and flexible characteristics and strong practicality.In the analysis of voice intelligence,the staff need to listen to the contents of a sentence to determine the value of a voice,cost a great deal of manpower and time.In this paper,based on the analysis of the characteristics of voice intelligence data,combined with the special needs of practical application,studies on the identification of phonetic information keywords.The main research work includes:1.For the particularity of voice intelligence in the voice data preprocessing,as far as possible use filtering,endpoint detection and other methods reasonably to eliminate all kinds of noise with complex environments in the pretreatment process,and by frequency compensation segment amplitude value normalized to ensure that the voice signal frequency moderate.2.Compare the influence of the MFCC and the first and second order differential components on the recognition effect,propose a method based on the MFCC weighting factor with reconstructing the feature parameters,according select the appropriate weights to improve the keyword recognition rate.3.In the recognition algorithm,the DTW algorithm is used to simulate the simulation parameters in the time domain and frequency domain,and the two algorithms are combined to propose a time domain frequency domain joint recognition algorithm.Compared with the results of the two algorithms,the joint algorithm effectively reduces the false alarm rate and improves the recognition rate.4.Implement a non-specific multi-lingual keyword recognition system,this system can select the audio file which need to be identified and can set one or more keywords,by searching the selected voice keyword,as a simulation of the initial analysis of intelligence information,the number of keywords and time to retrieve as a result of the identification will be displayed and stored.5.Establish a small vocabulary keyword corpus with no training and can be added,deleted and change at any time,recording in real time in different circumstances,create new keywords and use it to retrieve,also can capture the keywords which need of in audio file,as a new keywords to recognize the voice,which greatly improving the system practicality.