论文信息 - String and lattice based discriminative training for the corpus of spontaneous Japanese lecture transcription task

String and lattice based discriminative training for the corpus of spontaneous Japanese lecture transcription task

Abstract This article aims to provide a comprehensive set of acousticmodel discriminative training results for the Corpus of Spon-taneous Japanese (CSJ) lecture speech transcription task. Dis-criminativetrainingwascarriedoutforthistaskusinga100,000word trigram for several acoustic model topologies, using bothdiagonal and full covariance models, and using both string-based and lattice-based training paradigms. We describe ourimplementation of the proposal by Macherey et al. for numer-ical subtraction of the reference lattice statistics from the com-petitor lattice statistics during lattice-based Minimum Classiﬁ-cation Error (MCE) training. We also present results for lattice-based training that does not use such subtraction, correspond-ing to the well-known Maximum Mutual Information (MMI)approach. Discriminative training yielded relative reductionsin Word Error Rate of up to 13%. Speciﬁc problems encoun-tered in implementing discriminative training for this task arediscussed. 1. Introduction

Atsushi Nakamura | Erik McDermott

[1] Jonathan Le Roux,et al. Discriminative Training for Large-Vocabulary Speech Recognition Using Minimum Classification Error , 2007, IEEE Transactions on Audio, Speech, and Language Processing.

[2] Atsushi Nakamura,et al. Generalized fast on-the-fly composition algorithm for WFST-based speech recognition , 2005, INTERSPEECH.

[3] Fernando Pereira,et al. Weighted finite-state transducers in speech recognition , 2002, Comput. Speech Lang..

[4] Alex Acero,et al. Joint Discriminative Front End and Back End Training for Improved Speech Recognition Accuracy , 2006, 2006 IEEE International Conference on Acoustics Speech and Signal Processing Proceedings.

[5] Tatsuya Kawahara. Benchmark test for speech recognition using the Corpus of Spontaneous Japanese , 2003 .

[6] Daniel Povey,et al. Large scale discriminative training of hidden Markov models for speech recognition , 2002, Comput. Speech Lang..

[7] Daniel Povey. SPAM and full covariance for speech recognition , 2006, INTERSPEECH.

[8] Hitoshi Isahara,et al. Spontaneous Speech Corpus of Japanese , 2000, LREC.

[9] Hermann Ney,et al. Investigations on error minimizing training criteria for discriminative training in automatic speech recognition , 2005, INTERSPEECH.