CHAPTER 10 Applications of Digital Signal Processing Wang Weilian wlwang@ynu.edu.cn School of Information Science and Technology Yunnan University
Outline Speech Signals Processing Dual-Tone Multifrequency Signal Detection 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 2
Speech Analysis parameterize the speech signal To reduce the bandwidth To characterize the speech signal with only a few features Speech Signal Processing is one of the kernel technologies in those fields as follows: Information Superhighway, Multimedia, OAS (office automation system), Modern Communications System, Intelligent System and so on. 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 3
Speech Analysis time-domain data windowing: s ( n) s( m) w( n m) w m Windowing Calculation 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 4
Speech Analysis time-domain Energy: nn1 2 2 E [ x( m) w( n m)] [ x( m) w( n m)] n m mn 2 2 2 2 E [ x ( m) w ( n m)] [ x ( m) h( n m)] x ( n) h( n) n m m 2 h( n) w ( n) Selecting 10ms ~ 30ms as the length of the window in general 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 5
Speech Analysis time-domain The zero crossing rate ( ZCR ): Z sgn[ x( n)] sgn[ x( n 1)] w( n m) n m sgn[ x( n)] sgn[ x( n 1)] w( n) where 1 ( x 0) sgn[ x] 0 ( x 0) 1/ 2 N (0 n N 1) wn ( ) 0 other 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 6
Speech Analysis time-domain Energy and ZCR: Energy and ZCR 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 7
Speech Analysis time-domain The Autocorrelation function: Rn ( k ) [ x ( m ) w ( n m ) x ( m k ) w ( n ( m k ))] m R ( k) R ( k) [ x( m) w( n m) x( m k) w( n ( m k))] n n m If hk ( n ) w ( n ) w ( n k ), then: R ( k) [ x( m) x( m k)] h ( n m) [ x( n) x( n k)] h ( n) n k k m 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 8
Speech Analysis time-domain The Autocorrelation function: The block diagram of the autocorrelation function 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 9
Speech Analysis frequency-domain Fourier Transform and Spectrogram j X ( e ) x( m) w( n m) e n m jm The filter-explanation of the FT 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 10
Speech Analysis frequency-domain The spectrogram: The spectrogram 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 11
Speech Analysis frequency-domain Spectra analysis The power spectra ( energy density function ): j j j j n n n n S ( e ) X ( e ) X ( e ) X ( e ) Complex Ceptrum: 1 x( n) Z [ln[ Z[ x( n)]]] 2 Quefrency: [ x( n) x( n)] cn ( ) 2 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 12
Speech Analysis frequency-domain Linear Predictive: The model of signal generation Autoregressive Moving Average Model 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 13
Speech Analysis frequency-domain Linear Predictive: 1 H ( z) G is the gain factor. 1 q j1 p i1 bz j i j G az i Sz ( ) U( z) U(z) / S(z) is the Z-Transform of input / output sequence p q i j, i1 j0 s( n) a s( n i) G b u( n j) ( b 1) 0 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 14
Speech Analysis frequency-domain Linear Predictive ( autocorrelation method ) 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 15
Speech Synthesis Formant Synthesis: the transfer function of formants can be simulated by using a 2th-order digital filter generally. y( n) Ax( n) By( n 1) Cy( n 2) C exp( 2 B T ), B 2exp( B T)cos(2 FT ), A 1 B C w F is the formant frequency, and Bw is the bandwidth. w 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 16
Speech Recognition Communication via Spoken Language 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 17
Dual-Tone Multifrequency Signal Detection A DTMF signal consist of a sum of two tones with frequencies taken from two mutually exclusive groups of preassigned frequencies. Each pair of such tones represents a unique number or a symbol. 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 18
Dual-Tone Multifrequency Signal Detection Decoding of a DTMF signal thus involves identifying the two tones in that signal and determining their corresponding number or symbol. The DTMF decoder computes the DFT samples closest in frequency to the eight DTMF fundamental tones and their respective second harmonics. The DFT length N determines the frequency spacing between the locations of the DFT samples and the time it takes to compute the DFT sample. The frequency corresponding to the DFT index ( bin number ) k is: f kf N, k 0,1,, N 1 k T 云南大学滇池学院课程 : 数字信号处理 Applications of Digital Signal Processing 19