Constructing ensembles of ASR systems using randomized decision trees

Building multiple automatic speech recognition (ASR) systems and combining their outputs using voting techniques such as ROVER is an effective technique for lowering the overall word error rate. A successful system combination approach requires the construction of multiple systems with complementary errors, or the combination will not outperform any of the individual systems. In general, this is achieved empirically, for example by building systems on different input features. In this paper, we present a systematic approach for building multiple ASR systems in which the decision tree state-tying procedure that is used to specify context-dependent acoustic models is randomized. Experiments carried out on two large vocabulary recognition tasks, MALACH and DARPA EARS, illustrate the effectiveness of the approach.

[1]  Holger Schwenk,et al.  Using boosting to improve a hybrid HMM/neural network speech recognizer , 1999, 1999 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings. ICASSP99 (Cat. No.99CH36258).

[2]  Leo Breiman,et al.  Random Forests , 2001, Machine Learning.

[3]  Thomas G. Dietterich.  Multiple Classifier Systems , 2000, Lecture Notes in Computer Science.

[4]  Samy Bengio,et al.  Boosting HMMs with an application to speech recognition , 2004, 2004 IEEE International Conference on Acoustics, Speech, and Signal Processing.

[5]  Leo Breiman,et al.  Bagging Predictors , 1996, Machine Learning.

[6]  Yoav Freund,et al.  Experiments with a New Boosting Algorithm , 1996, ICML.

[7]  Olivier Siohan,et al.  Multiple classifiers by constrained minimization , 2000, 2000 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings (Cat. No.00CH37100).

[8]  L. Breiman.  Random Forests--random Features , 1999 .

[9]  Thomas G. Dietterich.  An Experimental Comparison of Three Methods for Constructing Ensembles of Decision Trees: Bagging, Boosting, and Randomization , 2000, Machine Learning.

[10]  Richard M. Schwartz,et al.  A compact model for speaker-adaptive training , 1996, Proceeding of Fourth International Conference on Spoken Language Processing. ICSLP '96.

[11]  Tin Kam Ho,et al.  The Random Subspace Method for Constructing Decision Forests , 1998, IEEE Trans. Pattern Anal. Mach. Intell..

[12]  Hermann Ney,et al.  Bootstrap estimates for confidence intervals in ASR performance evaluation , 2004, 2004 IEEE International Conference on Acoustics, Speech, and Signal Processing.

[13]  Zhou Zhi,et al.  Neural Network Ensemble , 2002 .

[14]  Lars Kai Hansen,et al.  Neural Network Ensembles , 1990, IEEE Trans. Pattern Anal. Mach. Intell..

[15]  Michael Picheny,et al.  Decision trees for phonological rules in continuous speech , 1991, [Proceedings] ICASSP 91: 1991 International Conference on Acoustics, Speech, and Signal Processing.

[16]  Bhuvana Ramabhadran,et al.  Automatic recognition of spontaneous speech for access to multilingual oral history archives , 2004, IEEE Transactions on Speech and Audio Processing.

[17]  Geoffrey Zweig,et al.  Boosting Gaussian mixtures in an LVCSR system , 2000, 2000 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings (Cat. No.00CH37100).

[18]  Wu Chou,et al.  Robust decision tree state tying for continuous speech recognition , 2000, IEEE Trans. Speech Audio Process..

[19]  Geoffrey Zweig,et al.  The IBM 2004 conversational telephony system for rich transcription , 2005, Proceedings. (ICASSP '05). IEEE International Conference on Acoustics, Speech, and Signal Processing, 2005..

[20]  Jonathan G. Fiscus,et al.  A post-processing system to yield reduced word error rates: Recognizer Output Voting Error Reduction (ROVER) , 1997, 1997 IEEE Workshop on Automatic Speech Recognition and Understanding Proceedings.