C. Wooters, 1993. "Lexical Modeling in a Speaker-Independent Speech Understanding System." Ph.D. Dissertation, University of California at Berkeley. Berkeley, CA. 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....average of 71 ms phone, while males had an average of 68 ms phone, a difference of about 4 , quite correlated with the similar differences in reduction and flapping. 4 Related Work Our algorithm for phonological rule probability estimation synthesizes and extends earlier work by (Cohen 1989) and (Wooters 1993). The idea of using optional phonological rules to construct a speechrecognition lexicon derives from Cohen (1989) who applied optional phonological rules to a baseform dictionary to produce a surface lexicon and then used TIMIT to assign probabilities for each pronunciation. The use of a ....

....lexicon derives from Cohen (1989) who applied optional phonological rules to a baseform dictionary to produce a surface lexicon and then used TIMIT to assign probabilities for each pronunciation. The use of a forced Viterbi speech decoder to discover pronunciations from a corpus was proposed by Wooters (1993). Wesenick Schiel (1994) independently propose a very similar forced Viterbidecoder based technique which they use for measuring the accuracy of hand written phonology. 6 Name Code Rule Pr Reductions Mid vowels RV1 stress [aa ae ah ao eh er ey ow uh] ax .60 High vowels RV2 stress [iy ih ....

Wooters, Charles C., 1993. Lexical Modeling in a Speaker Independent Speech Understanding System. Berkeley: University of California dissertation. Available as ICSI TR-92-062.

....with the output produced by the natural language on perfect strings, to get an idea for the semantic performance of the recognizer. Semantic Error Rate BeRP system 34.1 Backend alone 18.1 Recognizer alone 27.7 Table 7: BeRP semantic performance Further details of the BeRP system are presented in Wooters (1993) and Jurafsky et al. 1994) Acknowledgments We d like to thank Jerry Feldman for his early encouragement and support of this research, and Eric Fosler, Steve Omohundro, and an anonymous reviewer for extremely useful comments on this paper and the BeRP project. This work was partially funded by ....

WOOTERS, CHARLES C., 1993. Lexical Modeling in a Speaker Independent Speech Understanding System. Berkeley, CA: University of California dissertation.

....: 66 3.14 Number of HMM states as a function of the number of samples incorporated during incremental merging. 67 3. 15 Hybrid MLP HMM training merging procedure used in the BeRP speech understanding system (Wooters 1993). 73 4.1 Geometric and Poisson length distributions, for identical distribution mean (3.0) 87 4.2 Example grammars from Langley (1994) 101 7.1 Three ways of ....

....of merging with a small can be submitted to more merging at a larger value, until further increases reduce generalization on the cross validation data. CHAPTER 3. HIDDEN MARKOV MODELS 72 3.6. 3 Multiple pronunciation word models for speech recognition As part of his dissertation research, Wooters (1993) has used HMM merging extensively in the context of the Berkeley Restaurant Project (BeRP) BeRP is medium vocabulary, speaker independent spontaneous continuous speech understanding system that functions as a consultant for finding restaurants in the city of Berkeley, California (Jurafsky et al. ....

[Article contains additional citation context not shown here]

WOOTERS, CHARLES C., 1993. Lexical Modeling in a Speaker Independent Speech Understanding System.

....and feaspar use different input sentences, features and architectures. Besides parsec also the berp and trains systems focus on hybrid spoken language processing. berp (Berkeley Restaurant Project) is a current project which employs multiple different representations for speech language analysis (Wooters, 1993; Jurafsky et al. 1994, 1994b) The task of berp is to act as a knowledge consultant for giving advice about choos76 SCREEN: Flat Syntactic and Semantic Spoken Language Analysis ing restaurants. There are different components in berp: The feature extractor receives digitized acoustic data and ....

....spoken language analysis and to (2) the use of hybrid connectionist techniques which support the task by the choice of the possibly most appropriate knowledge structure. Many hybrid systems contain just a small portion of connectionist representations in addition to many other modules, e.g. berp (Wooters, 1993; Jurafsky et al. 1994, 1994b) janus (Waibel et al. 1992) trains (Allen, 1995; Allen et al. 1995) In contrast, most of the important subtasks in screen are performed directly by many connectionist networks. Furthermore, we have learned that flat syntactic and semantic representations could ....

Wooters, C. C. (1993). Lexical modeling in a speaker independent speech understanding system. Tech. rep. TR-93-068, International Computer Science Institute, Berkeley.

.... in the SICSA system, connectionist recurrent networks and symbolic case role parsing are combined for semantic analysis of database queries [4] In BerP, connectionist feedforward networks for speech processing and symbolic dialog understanding are combined in a spoken language dialog system [30, 65]. In FeasPar, connectionist feedforward networks for feature assignment are combined with symbolic search for finding better analysis results. As in the WP model, in all these loosely coupled hybrid processing architectures the connectionist module has to finish before the symbolic module can ....

C. C. Wooters. Lexical modeling in a speaker independent speech understanding system. Technical Report TR-93-068, International Computer Science Institute, Berkeley, 1993.

....including unification features and probabilistic verbal valence knowledge. We are also workingonimproving the efficiency of the tight coupling interface by adding an ungrammaticalprefix cache, obviating attempting to parse ungrammatical strings. Further details of the BeRP system are presented in Wooters (1993) and Jurafsky et al. 1994) Acknowledgments We d like to thank Jerry Feldman, Steve Omohundro,and Steve Renals for their generous help and advice. This work was partially funded by ICSI and an SRI subcontract from ARPA contract MDA904 90 C 5253. Partial funding also came from ESPRIT project ....

WOOTERS, CHARLES C., 1993. Lexical Modeling in a Speaker Independent Speech Understanding System. Berkeley, CA: University of California dissertation. available as ICSI TR-92-062.

....bigrams depending on the system s latest question. In very preliminary experiments with this discourse based tight coupling, we show a non significant 2 reduction in word error, but we are extremely optimistic that these results will improve. Further details of the BeRP system are presented in Wooters (1993) and Jurafsky et al. 1994) Acknowledgments This work was partially funded by ICSI and an SRI subcontract from ARPA contract MDA904 90 C 5253. Partial funding for the BeRP system development also came from ESPRIT project 6487 (The Wernicke project) 4. ....

WOOTERS, CHARLES C., 1993. Lexical Modeling in a Speaker Independent Speech Understanding System. Berkeley, CA: University of California dissertation. available as ICSI TR-92-062.

....average of 71 ms phone, while males had an average of 68 ms phone, a difference of about 4 , quite correlated with the similar differences in reduction and flapping. 4 Related Work Our algorithm for phonological rule probability estimation synthesizes and extends earlier work by (Cohen 1989) and (Wooters 1993). The idea of using optional phonological rules to construct a speech recognition 0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 80.00 90.00 100.00 FL1 FL2 RV1 RV2 RV3 SL1 SL2 SL3 SL4 VH1 timit wsj0 rec Rule Percent Percent of Phonological Rule Use, WSJ0 vs. TIMIT Figure 6: Automatic vs ....

....lexicon derives from Cohen (1989) who applied optional phonological rules to a baseform dictionary to produce a surface lexicon and then used TIMIT to assign probabilities for each pronunciation. The use of a forced Viterbi speech decoder to discover pronunciations from a corpus was proposed by Wooters (1993). Wesenick Schiel (1994) independently propose a very similar forced Viterbi decoder based technique which they use for measuring the accuracy of hand written phonology. Chen (1990) and Riley (1991) model the relationship between phonemes and their allophonic realizations by training decision ....

Wooters, Charles C., 1993. Lexical Modeling in a Speaker Independent Speech Understanding System.

....are flapped, etc. Despite the seemingly obvious advantage of MPMs, there has been conflicting evidence as to whether they can improve the performance of speech recognition systems. Some researchers [1] have not shown any improvements in recognition performance through the use of MPMs. Others [2, 3] have demonstrated significant improvements in performance. There are several factors that likely contribute to these conflicting reports. One such factor is the difficulty in constructing the MPMs. For example, how does one derive alternate pronunciations for a word How can we represent the ....

....implementation. Indeed, we have observed that the best results are obtained by combining the lexicon obtained in the first iteration with the MLP trained in the second iteration. 4 MODEL EVALUATION 4.1 Multiple vs. single pronunciations Previous experiments using the approach described here [3] addressed whether multiple pronunciation modeling in the BeRP system could give improvements over the traditional single pronunciationHMM word models. This comparison was based on a 2,319 training utterance corpus and a 364 utterance test set. The test set was gathered from 8 speakers, 4 males ....

[Article contains additional citation context not shown here]

C. Wooters. Lexical Modeling in a Speaker Independent Speech Understanding System. PhD thesis, University of California, Berkeley, 1993.

.... alone 18.1 Recognizer alone 27.7 Table 5: BeRP semantic performance We are currently extending the BeRP system to use visual data from the speakers lips, either to aid in recognition, or to help determine when a speaker has started speaking [3] Further details of the BeRP system are presented in [13] and [8] Acknowledgments Thanks to Emily Bender, Herve Bourlard, Jerry Feldman, Hana Filip, Hynek Hermansky, Ron Kay, Yochai Konig, Robert Moore, Steve Omohundro, Patti Price, and Liz Shriberg. This work was partially funded by ICSI, an SRI subcontract from ARPA contract MDA904 90 C 5253, and ....

Charles C. Wooters. Lexical Modeling in a Speaker Independent Speech Understanding System. PhD thesis, University of California, Berkeley, CA, 1993. available as ICSI TR-92-062.

No context found.

C. Wooters, 1993. "Lexical Modeling in a Speaker-Independent Speech Understanding System." Ph.D. Dissertation, University of California at Berkeley. Berkeley, CA. 1993.

No context found.

WOOTERS,CHARLES C., 1993. Lexical Modeling in a Speaker Independent Speech Understanding System. Berkeley, CA: University of California dissertation.

No context found.

C.C. Wooters, Lexical Modeling in a Speaker Independent Speech Understanding System, U.C. Berkeley Ph.D. Thesis, 1993. 36

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC