id	author	title	date	pages	extension	mime	words	sentences	flesch	summary	cache	txt
work_pa2cgjgwrnepbdm4hnu6xpauc4	Chia-ying Lee	Unsupervised Lexicon Discovery from Acoustic Input	2015	16	.pdf	application/pdf	9347	823	62	model builds on earlier models of unsupervised phone-like unit discovery from acoustic data (Lee and Glass, 2012), and unsupervised symbolic lexicon discovery using the on earlier work addressing the unsupervised discovery of phone-like units from acoustic data—in particular the Dirichlet Process Hidden Markov Model We provide preliminary evidence that simultaneously learning sound and lexical structure leads to synergistic interactions (Johnson, 2008b)—the various components of the model here are models which treat segmentation as a secondary consequence of discovering a compact lexicon which explains the distribution of phoneme sequences in the input (Cartwright and Brent, 1994; a fixed, two-level representation of linguistic structure, our use of adaptor grammars to model symbolic lexicon discovery means that we can easily and Phonological variability modeling In conversational speech, the phonetic realization of a word can In summary, our model integrates adaptor grammars with a noisy-channel model of phonetic variability and an acoustic model to discover hierarchical linguistic structures directly from acoustic signals.	./cache/work_pa2cgjgwrnepbdm4hnu6xpauc4.pdf	./txt/work_pa2cgjgwrnepbdm4hnu6xpauc4.txt