课程名称︰数位语音处理概论 课程性质︰选修 课程教师︰李琳山 开课学院：电资学院 开课系所︰资工系 考试日期（年月日）︰2010.5.19 考试时限（分钟）：120 是否需发放奖励金：是 （如未明确表示，则不予发放） 试题 : #OPEN Lecture Power Point (Printed Version) and Personal Notes #You have to use CHINESE sentences to answer all the questions, but you can use English terminologies #Total points: 160 ------------------------------------------------------------------------------ _ 1.(20)Given a GMM λ=(A,B,π)with N states, an observation sequence O=o1o2...ot ...oT αt(i) = Prob[o1o2...ot,qt = i | λ] βt(i) = Prob[o[t+1]o[t+2]...oT | qt = i, λ] N (a)Derive the meaning ofΣ αt(i)βt(i) in probability. i-1 αt(i)βt(i) (b)Derive the meaning of --------------- in probability. N Σ αt(j)βt(j) j=1 (c)Derive the meaning of αt(i)a[ij]bj(o[t+1])β[t+1](j) in probability. (d)Formulate and describe the Viterbi algorithm to find the best state sequence _* * * * * _ _* q = q1q2...qt...qT giving the highest probability Prob[O, q | λ]. Why do we need backtracking? 2.(20)What is the problem of coarticulation and context dependency considered in acoustic modeling? Describe a situation that mono-phone model outperforms tri-phone model. 3.(20)In large vocabulary continuous speech recognition, explain: (a)What the "language model weight" is. (b)Why the language model has the function as the penalty of inserting extra words. 4.(20)Explain what the class-based language model is. What can be the possible reasons if class-based N-gram model perform worse than word-based N-gram model? 5.(20)What is the perplexity of a language source? What is the perplexity of a language model with respect to a corpus? How are they related to a "virtual vocabulary"? 6.(20)In feature extraction for speech recognition, after you obtain 12 MFCC parameters plus a short-time energy (a total of 13 parameters), explain how to obtain the other 26 parameters and what they are. 7.(20)In training HMM models for isolated word recognition, do you think the more number of iterations you perform, the higher recognition accuracy you will get? Note that it is guaranteed that the likelihood will be increased in each iteration. Explain your answer. 8.(20)What is the pre-emphasis procedure in MFCC extraction? Why do we perform pre-emphasis? --

※ 发信站: 批踢踢实业坊(ptt.cc) ◆ From: 140.112.30.96