2.5.4.1 Basics of Neural Networks X0 X1 INPUT Y OUTPUT N 1 y f Wi xi i 0 X2 X N 1 1 2.5.4.2 Neural Network Topologies 2 2.5.4.2 Neural Network Topologies 3 2.5.4.2 Neural Network Topologies 4 TDNN 5 2.5.4.6 Neural Network Structures for Speech Recognition 6 2.5.4.6 Neural Network Structures for Speech Recognition 7 3.1.1 Spectral Analysis Models 8 3.1.1 Spectral Analysis Models 9 3.2 THE BANK-OF-FILTERS FRONT- END PROCESSOR 10 3.2 THE BANK-OF-FILTERS FRONT- END PROCESSOR 11 3.2 THE BANK-OF-FILTERS FRONT- END PROCESSOR 12 3.2 THE BANK-OF-FILTERS FRONT- END PROCESSOR 13 3.2 THE BANK-OF-FILTERS FRONT- END PROCESSOR 14 3.2.1 Types of Filter Bank Used for Speech Recognition Fs f i i, N Q N /2 1 i Q Fs bi N 15 Nonuniform Filter Banks b1 c bi bi 1 , 2iQ (bi b1 ) f i f1 b j , 2 j 1 i 1 16 Nonuniform Filter Banks Filter 1 : f1 300 Hz , b1 200 Hz Filter 2 : f 2 600 Hz , b2 400 Hz Filter 3 : f 3 1200 Hz , b3 800 Hz Filter 4 : f 4 2400 Hz , b4 1600 Hz 17 3.2.1 Types of Filter Bank Used for Speech Recognition 18 3.2.1 Types of Filter Bank Used for Speech Recognition 19 3.2.2 Implementations of Filter Banks Instead of direct convolution, which is computationally expensive, we assume each bandpass filter impulse response to be represented by: hi (n) w(n)e j i n Where w(n) is a fixed lowpass filter 20 3.2.2 Implementations of Filter Banks 21 3.2.2.1 Frequency Domain Interpretation of the ShortTime Fourier Transform 22 3.2.2.1 Frequency Domain Interpretation of the Short-Time Fourier Transform 23 3.2.2.1 Frequency Domain Interpretation of the Short-Time Fourier Transform 24 3.2.2.1 Frequency Domain Interpretation of the Short-Time Fourier Transform 25 Linear Filter Interpretation of the STFT ~ s ( n) s (n) w(n) e S n (e j i ji 26 ) 3.2.2.4 FFT Implementation of a Uniform Filter Bank 27 Direct implementation of an arbitrary filter bank s (n) h1 (n) X 1 ( n) h2 (n) X 2 (n) hQ (n) X Q (n) 28 3.2.2.5 Nonuniform FIR Filter Bank Implementations 29 3.2.2.7 Tree Structure Realizations of Nonuniform Filter Banks 30 3.2.4 Practical Examples of SpeechRecognition Filter Banks 31 3.2.4 Practical Examples of SpeechRecognition Filter Banks 32 3.2.4 Practical Examples of SpeechRecognition Filter Banks 33 3.2.4 Practical Examples of SpeechRecognition Filter Banks 34 3.2.5 Generalizations of Filter-Bank Analyzer 35 3.2.5 Generalizations of Filter-Bank Analyzer 36 3.2.5 Generalizations of Filter-Bank Analyzer 37 3.2.5 Generalizations of Filter-Bank Analyzer 38 40 41 43 44 روش MFCC روش MFCCمبتني بر نحوه ادراک گوش انسان از اصوات مي باشد. روش MFCCنسبت به ساير وي ِژگيها در محيطهاي نويزي بهتر عمل ميکند. ً MFCCاساسا جهت کاربردهاي شناسايي گفتار ارايه شده است اما در شناسايي گوينده نيز راندمان مناسبي دارد. واحد شنيدار گوش انسان Melمي باشد که به کمک رابطه زير بدست مي آيد: 45 مراحل روش MFCC مرحله :1نگاشت سيگنال از حوزه زمان به حوزه فرکانس به کمک FFTزمان کوتاه. ) :Z(nسيگنال گفتار ( :W(nتابع پنجره مانند پنجره همينگ WF= e-j2π/F ;m : 0,…,F – 1 :Fطول فريم گفتاري. 46 مراحل روش MFCC مرحله :2يافتن انرژي هر کانال بانک فيلتر. که Mتعداد بانکهاي فيلتر مبتني بر معيار مل ميباشد. تابعkفيلترهاي بانک فيلتر است. 0,1,..., M 1 ) Wk ( j 47 توزيع فيلترمبتنی برمعيار مل 48 مراحل روش MFCC 49 مرحله :4فشرده سازي طيف و اعمال تبديل DCTجهت حصول به ضرايب MFCC در رابطه باال n=0،...،Lمرتبه ضرايب MFCCميباشد. کپستروم-روش مل سیگنال زمانی فریم بندی |FFT|2 Mel-scaling Logarithm IDCT Cepstra Delta & Delta Delta Cepstra Differentiator Low-order coefficients 50 Time-Frequency analysis Short-term Fourier Transform Standard way of frequency analysis: decompose the incoming signal into the constituent frequency components. W(n): windowing function N: frame length p: step size 51 Critical band integration Related to masking phenomenon: the threshold of a sinusoid is elevated when its frequency is close to the center frequency of a narrow-band noise Frequency components within a critical band are not resolved. Auditory system interprets the signals within a critical band as a whole 52 Bark scale 53 Feature orthogonalization Spectral values in adjacent frequency channels are highly correlated The correlation results in a Gaussian model with lots of parameters: have to estimate all the elements of the covariance matrix Decorrelation is useful to improve the parameter estimation. 54