论文信息 - Analysing Recognition Errors in Unlimited-Vocabulary Speech Recognition

Analysing Recognition Errors in Unlimited-Vocabulary Speech Recognition

We analyze the recognition errors made by a morph-based continuous speech recognition system, which practically allows an unlimited vocabulary. Examining the role of the acoustic and language models in erroneous regions shows how speaker adaptive training (SAT) and discriminative training with minimum phone frame error (MPFE) criterion decrease errors in different error classes. Analyzing the errors with respect to word frequencies and manually classified error types reveals the most potential areas for improving the system.

Mikko Kurimo | Teemu Hirsimäki

[1] Richard M. Schwartz,et al. Analysis of the errors produced by the 2004 BBN speech recognition system in the DARPA EARS evaluations , 2006, IEEE Transactions on Audio, Speech, and Language Processing.

[2] Teemu Hirsimäki,et al. On Growing and Pruning Kneser–Ney Smoothed $ N$-Gram Models , 2007, IEEE Transactions on Audio, Speech, and Language Processing.

[3] Mark J. F. Gales,et al. Maximum likelihood linear transformations for HMM-based speech recognition , 1998, Comput. Speech Lang..

[4] Mikko Kurimo,et al. Importance of High-Order N-Gram Models in Morph-Based Speech Recognition , 2009, IEEE Transactions on Audio, Speech, and Language Processing.

[5] Lin Lawrance Chase. Error-responsive feedback mechanisms for speech recognizers , 1997 .

[6] Andreas Stolcke,et al. Improved discriminative training using phone lattices , 2005, INTERSPEECH.

[7] Steven Greenberg,et al. AN INTRODUCTION TO THE DIAGNOSTIC EVALUATION OF SWITCHBOARD-CORPUS AUTOMATIC SPEECH RECOGNITION SYSTEMS , 2000 .

[8] Mikko Kurimo,et al. Unlimited vocabulary speech recognition with morph language models applied to Finnish , 2006, Comput. Speech Lang..