Search Strategies For Large-Vocabulary Continuous-Speech Recognition

Ney, Hermann

doi:10.1007/978-3-642-57745-1_29

Cited by 14 publications

(12 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For a bigram language model [3], a separate copy of the lexical tree is needed for each predecessor word v. When going from a bigram to a trigram language model, we have to take into account that, for a trigram language model, the probabilities are conditioned on the previous two predecessor words (u; v) rather than one predecessor word v in the bigram case [6,7]. Therefore, we have to make the copies dependent on the two predecessor words.…”

Section: Word Conditioned Searchmentioning

confidence: 99%

“…To describe the time conditioned search algorithm, we define the following quantities as introduced in [6]: h(w; ;t)= probability that word w produces the acoustic vectors x+1:::xt. H(v; ) = probability that the acoustic vectors x1:::x are generated by a word/state sequence with v as the last word and as the word boundary.…”

Section: Bigram Language Modelsmentioning

confidence: 99%

“…The paper describes two variants of the time-synchronous beamsearch algorithm using a tree-organized pronunciation lexicon [6]. Both methods are based on a one-pass concept and utilize a (prefix) tree organization of the pronunciation lexicon [3].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A comparison of time conditioned and word conditioned search techniques for large vocabulary speech recognition

Ortmanns

Ney

Seide³

et al.

Proceeding of Fourth International Conference on Spoken Language Processing. ICSLP '96

View full text Add to dashboard Cite

In this paper, we compare the search effort of the word conditioned and the time conditioned tree search methods. Both methods are based on a time-synchronous, left-to-right beam search using a treeorganized lexicon. Whereas the word conditioned method is well known and widely used, the time conditioned method is novel in the context of 20 000-word vocabulary recognition. We extend both methods to handle trigram language models in a one-pass strategy. Both methods were tested on a train schedule inquiry task(1 850 words, telephone speech) and on the North American Business (Nov.'94) development corpus (20 000 words).

show abstract

Section: Word Conditioned Searchmentioning

confidence: 99%

Section: Bigram Language Modelsmentioning

confidence: 99%

See 1 more Smart Citation

A comparison of time conditioned and word conditioned search techniques for large vocabulary speech recognition

Ortmanns

Ney

Seide³

et al.

Proceeding of Fourth International Conference on Spoken Language Processing. ICSLP '96

View full text Add to dashboard Cite

show abstract

“…The first estimation method considered, here called Linear Simple (LS), assumes a simple discounting constant that can be estimated by assuming either a Poisson process for new words occurring after a given context (Witten & Bell, 1991), or by applying the LOO estimation method (Nadas, 1985;Ney et al, 1994). In both cases a good approximation of the so computed estimators yields the GT estimator for novel bigrams:…”

Section: Good-turing (Gt) Formulamentioning

confidence: 99%

“…Another advantage is that a significantly smaller amount of n-grams have to be kept in storage, as most n-grams in real texts occur once or twice. By assuming instead 0< <1 and by applying the Leaving-one-out (LOO) estimation criterion, a different solution (S ) was provided by Ney et al (1994) (see Table II). The same authors also claimed that no improvements were seen by assuming as a function of the context y.…”

Section: Good-turing (Gt) Formulamentioning

confidence: 99%