Font Size: a A A

Acoustic Modelling For Local Accent Under Mandarin Study

Posted on:2007-02-20Degree:MasterType:Thesis
Country:ChinaCandidate:S Y HuangFull Text:PDF
GTID:2178360185474577Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Along with raise continuously of mandarin usage scope and request, amount and difficulty of mandarin study and training work present an index number growth, expanding mandarin to work still very difficult.huge.In the meantime along with the technical development of speech recognition, speech recognition technique has already make so great progress that it is possible make use of speech recognition technique to carry out mandarin study and training thus.Mandarin 's studying and training system generally and all is according to speech recognition system, it often recognise the inaccurate likeness, misty of pronounce being subjected to Sichuan-accent influence as right.So we have t the necessity to carry on a valid distinction for the speech be influenced by local accent, in order to raise the function of mandarin study and train system.Textual and main contents is acoustic modelling towards those speech being subjected to Sichuan-accent influence and extraction the sensitive characteristic parameters s for Sichuan-accent;Discussed the usefulness of this kind of characteristic parameters s for the speech being subjected to Sichuan-accent influence.Firstly, according to the characteristics of Sichuan-accented mandarin ,the paper analyse differentiation of the standard mandarin and the Sichuan-accented mandarin and choose the speech whose initial consonants are zh,z,ch,c,sh and s as experimental object.Secondly, in consideration of the difficultie of speech endpoint detection under low SNR, there brings forward a method of endpoint detection based on space of higher dimension corner of auto-Correlated.The complexity of this method is lower than some currently algorithm and can detect effectively speech endpoint.Again, for the sake of recognising efficiently the likeness, misty of pronounce being subjected to Sichuan-accent influence while carring on a mandarin study and training, this paper carry on a acoustic modelling for the speech being subjected to Sichuan-accent influence and extraction the sensitive characteristic parameters for Sichuan-accent.Concretely will divide two steps to carry on: extracting original feature parameters s consisted of LPCC, difference LPCC,MFCC and difference MFCC of 12 dimensions and pitch; extract characteristic parameters s subjected to Sichuan-accent influence easily based on the orthogonal design,adding and reduceing dimensions,weighted features.
Keywords/Search Tags:Speech recognition, Local accent, Corner of auto-correlated, Secondary characteristic parameters
PDF Full Text Request
Related items