Analysis and implementation of the speaker adaptation techniques : MAP, MLLR, and MLED

[摘要] ENGLISH ABSTRACT:The topic of this thesis is speaker adaptation, whereby speaker-independent speech modelsare adapted to more closely match individual speakers by utilising a small amount ofdata from the targeted individual. Speaker adaptation methods - specifically, the MAP,MLLR and MLED speaker adaptation methods - are critically evaluated and compared.Two novel extensions of the MLED adaptation method are introduced, derived andevaluated. The first incorporates the explicit modelling of the mean speaker model inthe speaker-space into the MLED framework. The second extends MLED to use basisvectors modelling inter-class variance for classes of speech models, instead of basis vectorsmodelling inter-speaker variance.An evaluation of the effect of two different types of feature vector - PLP-cepstra andLPCCs - on the performance of speaker adaptation is made, to determine which featurevector is optimal for speaker-independent systems and the adaptation thereof.

[发布机构] Stellenbosch University