Speech recognition for DARPA Communicator

Aaron, A.; Chen, S.; Cohen, Paul S.; Dharanipragada, S.; Eide, Ellen; Franz, Martin; LeRoux, J.-M.; Luo, Xiaoqiang; Maison, B.; Mangu, Lidia; Mathes, Timothy K.; Novák, Miroslav; Olsen, Peder A.; Picheny, Michael; Printz, Harry; Ramabhadran, Bhuvana; Sakrajda, Andrej; Saon, George; Tydlitat, B.; Visweswariah, Karthik; Yuk, Dong-Suk

doi:10.1109/icassp.2001.940874

Cited by 7 publications

(1 citation statement)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In contrast, we next interpolated the state-independent class trigram with a state-dependent class trigram. As in [3], we found that estimating weights chosen to optimize the language model perplexity of a development corpus was worse than simply using i in all the states, perhaps due to insufficient development data. Table 2 shows that the word error rate, averaged over states 2 through 4, was reduced from 21.1% to 20.6%, a relative reduction of 2.4% (adapted 3-gram).…”

Section: ¢ -Gramsmentioning

confidence: 58%

Adaptive language models for spoken dialogue systems

Solsona¹,

Fosler-Lussier²,

Kuo³

et al. 2002

IEEE International Conference on Acoustics Speech and Signal Processing

View full text Add to dashboard Cite

In this paper, we investigate both generative and statistical approaches for language modeling in spoken dialogue systems. Semantic class-based finite state and ¢ -gram grammars are used for improving coverage and modeling accuracy when little training data is available. We have implemented dialogue-state specific language model adaptation to reduce perplexity and improve the efficiency of grammars for spoken dialogue systems. A novel algorithm for combining state-independent ¢ -gram and state-dependent finite state grammars using acoustic confidence scores is proposed. Using this combination strategy, a relative word error reduction of 12% is achieved for certain dialogue states within a travel reservation task. Finally, semantic class multigrams are proposed and briefly evaluated for language modeling in dialogue systems.

show abstract

Section: ¢ -Gramsmentioning

confidence: 58%