Rejection techniques based on context independent subword units

Torrecilla, Juan Carlos; Merino, Daniel Tapias; Caminero-Gil, F. Javier; Villarrubia, Luis

doi:10.21437/eurospeech.1995-395

Cited by 2 publications

(2 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Now we will test our method in another different task, that is, an isolated word recognition task based on context-dependent units. In these systems, we previously used a garbage model constructed as a combination of context independent subword HMMs trained using only the keyword training set, as we presented in [6]. In Table 3, we present the comparison between explicit garbage techniques and on-line rejection techniques, in a vocabulary-independent isolated recognition task of spanish surnames.…”

Section: Resultsmentioning

confidence: 99%

“…For this "traditional" verification scheme we have been using different explicit garbage models depending on the particular recognition task. That is, for natural number recognition we use a four-state left-to-right HMM, while for vocabularyindependent recognition we use a net of contextindependent units in parallel [6]. Following this scheme an utterance is rejected when the garbage model provides a garbage score or garbage likelihood which is better than any of the vocabulary words.…”

Section: On-line Garbage Modelingmentioning

confidence: 99%

See 1 more Smart Citation

On-line garbage modeling with discriminant analysis for utterance verification

Caminero-Gil¹,

Torre²,

Villarrubia³

et al. 1996

4th International Conference on Spoken Language Processing (ICSLP 1996)

View full text Add to dashboard Cite

Out-of-vocabulary (OOV) utterance detection and rejection are specially important and difficult problems in large-vocabulary and continuous speech recognition. In [1] we proposed an utterance verification procedure based on the use of frame-by-frame best acoustic state scores instead of using explicit garbage models. This procedure is usually referred to as on-line garbage modeling.In this contribution we extend our previous work in two major directions: a) we analyze, through the use of Discriminant Analysis, the possibilities of using L-best local scores and N-best utterance hypotheses scores for utterance verification; b) we present experimental results not only for a spontaneously spoken natural number recognition task, as in [1], but also for a flexible large vocabulary recognition task. All the results, based on a telephone database, show that the proposed on-line garbage modeling procedure outperforms, both in performance and computational cost, to other approaches based on the use of explicit garbage models.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: On-line Garbage Modelingmentioning

confidence: 99%