Language_Modeling_for_Speech_Recognition

Language_Modeling_for_Speech_Recognition - Speech...

Info iconThis preview shows pages 1–11. Sign up to view the full content.

View Full Document Right Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Speech Recognition Language Modeling for Speech Recognition February 13, 2012 Veton Kpuska 2 Language Modeling for Speech Recognition Introduction n-gram language models Probability estimation Evaluation Beyond n-grams February 13, 2012 Veton Kpuska 3 Language Modeling for Speech Recognition Speech recognizers seek the word sequence which is most likely to be produced from acoustic evidence A. Speech recognition involves acoustic processing, acoustic modeling, language modeling, and search Language models (LMs) assign a probability estimate P ( W ) to word sequences W = { w 1 ,...,w n } subject to ( 29 ( 29 ( 29 ( 29 W P W A P A W P A W P W W | max | max | = ( 29 1 = W W P February 13, 2012 Veton Kpuska 4 Language Modeling for Speech Recognition Language models help guide and constrain the search among alternative word hypotheses during recognition February 13, 2012 Veton Kpuska 5 Language Model Requirements February 13, 2012 Veton Kpuska 6 Finite-State Networks (FSN) Language space defined by a word network or graph. Describable by a regular phrase structure grammar. A a B | a Finite coverage can preset difficulties for ASR. Graph arcs or rules can be augmented with probabilities. February 13, 2012 Veton Kpuska 7 Context-Free Grammars (CFGs) Language space defined by context-free rewrite rules, e.g., A BC | a More powerful representation than FSNs Stochastic CFG rules have associated probabilities which can be learned automatically from corpus. Finite coverage can present difficulties for ASR. February 13, 2012 Veton Kpuska 8 Word-Pair Grammars & Bigrams Language space defined by lists of legal word-pairs Can be implemented efficiently within Viterbi search Finite coverage can present difficulties for ASR Bigrams define probabilities for all word-pairs and can produce a nonzero P ( W ) for all possible sentences February 13, 2012 Veton Kpuska 9 Example of LM Impact (Lee, 1988) Resource Management domain Speaker-independent, continuous speech corpus Sentences generated form a finite state network 997 word vocabulary Word-pair perplexity ~60. Bigram ~20 Error includes substitutions, deletions, and insertions February 13, 2012 Veton Kpuska 10 Note on Perplexity Language can be thought of as an information source whose outputs are words w i belonging to the vocabulary of the language....
View Full Document

Page1 / 55

Language_Modeling_for_Speech_Recognition - Speech...

This preview shows document pages 1 - 11. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online