Ninth International Conference on Spoken Language Processing

Pittsburgh, PA, USA
September 17-21, 2006

Time-Dependent Cross-Probability Model for Multi-Environment Model Based LInear Normalization

Luis Buera (1), Eduardo Lleida (1), Juan A. Nolazco-Flores (2), Antonio Miguel (1), Alfonso Ortega (1)

(1) Universidad de Zaragoza, Spain; (2) ITESM, Mexico

In a previous work, Multi-Environment Model based LInear Normalization, MEMLIN, was presented and it was proved to be effective to compensate environment mismatch. MEMLIN is an empirical feature vector normalization which models clean and noisy spaces by Gaussian Mixture Models (GMMs). In this algorithm, the probability of the clean model Gaussian, given the noisy model one and the noisy feature vector (cross-probability model) is a critical point. In the previous work the cross-model probability was approximated as time-independent. In this paper, a time-dependent estimation of the cross-probability model based on GMM is proposed. Some experiments with SpeechDat Car database were carried out in order to study the performance of the proposed estimation in a real acoustic environment. MEMLIN with time-independent cross-probability model reached 70.21% of mean improvement in Word Error Rate (WER), however, when timedependent cross-probability model based on GMM was applied, the mean improvement in WER went up to 78.47%.

Full Paper

Bibliographic reference.  Buera, Luis / Lleida, Eduardo / Nolazco-Flores, Juan A. / Miguel, Antonio / Ortega, Alfonso (2006): "Time-dependent cross-probability model for multi-environment model based LInear normalization", In INTERSPEECH-2006, paper 1271-Wed1BuP.2.