《语音线性预测编码课件》由会员分享,可在线阅读,更多相关《语音线性预测编码课件(32页珍藏版)》请在金锄头文库上搜索。
1、1,Speech Signal ProcessingLecture 10,Linear Predictive Coding (LPC)-Introduction,2,LPC Methods,LPC methods are the most widely used in speech coding, speech synthesis, speech recognition, speaker recognition and verification and for speech storage LPC methods provide extremely accurate estimates of
2、speech parameters, and does it extremely efficiently basic idea of Linear Prediction: current speech sample can be closely approximated as a linear combination of past samples, i.e.,3,LPC Methods,4,Basic Principles of LP,H(z),5,LP Basic Equations,6,LP Estimation Issues,7,Solution for k,8,Solution fo
3、r k,9,Solution for k,10,Solution for k,11,Autocorrelation Method,12,Autocorrelation Method,13,Autocorrelation Method,14,Covariance Method,15,Covariance Method,16,Cholesky Decomposition Example,17,Cholesky Decomposition,18,Cholesky Decomposition Example,19,Covariance Method Minimum Error,20,Solution
4、to Autocorrelation Method,21,Autocorrelation Example,22,Comparisons Between LP Methods,the various LP solution techniques can be compared in a number of ways, including the following: computational issues numerical issues stability of solution number of poles (order of predictor) window/section leng
5、th for analysis,23,LP Solution Computations,assume N1 N2p; choose values of N1=300, N2=300, p=10computation for covariance method N1p+p3 4000 *,+autocorrelation method N2p+p2 3100 *,+,24,LP Solution Comparisons,stability guaranteed for autocorrelation method cannot be guaranteed for covariance metho
6、d; as window size gets larger, this almost always makes the system stable choice of LP analysis parameters need 2 poles (for 1 kHz) for each vocal tract resonance below Fs/2 need 3-4 poles to represent source shape and radiation load use values of p 13-14,25,Prediction Error Signal Behavior,26,LP An
7、alysis of Speech,note error signal near time=0 for autocorrelation method,27,Frequency Domain Interpretation,28,Frequency Domain Interpretation,29,Frequency Domain Interpretation,30,Comparisons of Spectral Analysis Methods,Spectral Comparisons across Methodsa: narrowband analysisresolves pitch harmo
8、nicsb: wideband analysisgives a smoothed spectrumc: homomorphic smoothinggives good indications of formant locationsd: LPC fit with p=12formants and bandwidths can readily be obtained by factoring LPC polynomial,31,LPC Synthesis,32,Formant Analysis Using LPC,factor predictor polynomialassign roots to formantspick prominent peaks in LPC spectrumproblems on nasals where roots are not poles or zeros,