作者: Qi Li
DOI: 10.1007/978-3-642-23731-7_6
关键词: Sequential probability ratio test 、 Beam diameter 、 Path (graph theory) 、 Computer science 、 Speech recognition 、 Task (project management) 、 Speaker recognition 、 Hidden Markov model 、 Decoding methods 、 Viterbi algorithm
摘要: Decoding or searching is an important task in both speaker and speech recognition. In verification (SV), given a spoken password speakerdependent hidden Markov model (HMM), the of decoding to find optimal state alignments sense maximum likelihood score entire utterance. Currently, most popular algorithm Viterbi with pre-defined beam width reduce search space; however, it difficult determine suitable beforehand. A small may miss path while large one slow down process. To address problem, author has developed non-heuristic space. The details are presented this chapter.