Research On Whisper Speech Signal Processing And Its Application In Laser Interception

Laser interception is using a laser to obtain the voice signals,taking further analysis and processing of the signal.Laser interception is usually adopted for laser measurement technology which is the development of laser doppler technology.Utilizing laser measurement technology can realize non-contact,real-time,and accurate measurement of a variety of physical quantities,now the laser measurement technology is widely used in various fields,such as scientific research,industrial production,automatic control,analysis and measurement.As a kind of distinctive phonetic form in the daily life,the whisper speech can make people adapt to more occasions when communicating.Whisper speech always contains information that the speaker deliberately hides,combined with the laser measurement technology study on recognition of whisper speech,it is able to obtain non-contact,remote,secure secret access to high SNR speech signal.Meanwhile,further explore with the speaker,speech content and other valuable information can be taken through signal processing.It has enormous application value in the field of information security,network communication,automation,forensic identification and so on.This paper presents the working principle of Doppler laser vibrometer(LDV).And direct or indirect acquisition of ear speech signals by measuring the sound field directly or measuring the vibration of thin-walled objects.Along with investigating the noise signals processing,in order to get high SNR whisper speech.And the composition of a complete speaker recognition system is well explained.The specific steps of speaker recognition,and introduce several widely used speaker system models.Because of the vocal cords are not vibrating,the energy is low,and no fundamental frequency of the whisper,on each steps of speaker recognition,especially the selection of feature parameter extraction,more improvement is needed.This paper introduced and compared including linear prediction cepstral coefficients(LPCC),Mel frequency cepstrum coefficient(MFCC),and two more aimed at whisper improved MFCC feature parameter extraction.On the basis of this,an experiment of measuring the vibration of glass by laser to obtain the speech signal is taken,and using the traditional recording device to obtain the same ear voice signal is executed.Then Gauss mixture model(GMM)is selected as the speaker recognition model to carry out the speaker recognition experiment of two groups of speech signals.According to the experimental results,the recognition effect of laser collection whisper speech signal is displayed,comparing to the traditional way of getting a whisper speech.And the comparison of the recognitioneffect on different feature parameters extraction algorithm is made with the same whisper speech signal.