Font Size: a A A

Normalized Based On Conditional Random Spoken

Posted on:2010-04-14Degree:MasterType:Thesis
Country:ChinaCandidate:B XuFull Text:PDF
GTID:2208360275498895Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Natural manual-machine interface is the key technology which decides the computer can be used wildly or not.As a typical application of the natural manual-machine interface, Speech-to-speech translation has already made some achievements at present,but there are still a lot of questions to be studied further.In particular,because of the variability and flexibility of natural spoken languages,there are a large number of non-standardization phenomenons,such as repetition,redundance,ellipsis and so on.It is inevitable that there are recognition faulty results after speech recognition.These all have great influence on following treatment process of speech translation system,and make low-quality translation.This thesis is based on Conditional Random Fields model,and develops the research on clearing up non-standardization phenomenons,and rectifing faulty recognition results.The main work and characteristic are as follows:1.The paper introduces in detail the theory of Conditional Random Fields model and itsimpact on the natural language processing.We analyze and compare the advantages ofthe model compared with the other sequence tagging statistical models.2.Accordance to the problems which exist in Speech-to-speech translation system,we specifically carry out research on clearing up non-standardization phenomenon and rectifing faulty recognition results in natural spoken languages.3.In the paper,we use the combination of rules and statistical methods to deal with this issue.According to the characteristics of natural spoken language,this article summarized some of the language rules can be used.The introduction of rules solves a number of drawbacks based simply on statistical method.The experimental results show that this method is better than the way based simply on statistical method.4.We design and program the Conditional Random Fields Model.It solves the problem that the memory demand of available model tools is very large,and so that it improves efficiency of projects obviously.5.Using the available corpus,this paper designs an intact experiment system.The experimental results indicate that effect of oral standardization processing based on Conditional Random Fields model is good.
Keywords/Search Tags:Speech-to-speech Translation System, Oral Standardization Processing, Conditional Random Fields, Feather
PDF Full Text Request
Related items