Acoustical And Environmental Robustness In Automatic Speech Recognition Book

	List of Figures
	List of Tables
	Foreword
	Acknowledgments
1	Introduction	1
1.1	Acoustical Environmental Variability and its Consequences	2
1.2	Previous Research in Signal Processing for Robust Speech Recognition	5
1.3	Towards Environment-Independent Recognition	13
1.4	Monograph Outline	15
2	Experimental Procedure	17
2.1	An Overview of SPHINX	17
2.2	The Census Database	21
2.3	Objective Measurements	24
2.4	Baseline Recognition Accuracy	30
2.5	Other Databases	31
3	Frequency Domain Processing	39
3.1	Multi-Style Training	39
3.2	Channel Equalization	41
3.3	Noise Suppression by Spectral Subtraction	42
3.4	Experiments with Sphinx	48
4	The SDCN Algorithm	67
4.1	A Model of the Environment	67
4.2	Processing in the Frequency Domain: The MMSEN Algorithm	69
4.3	Processing in the Cepstral Domain: The SDCN Algorithm	74
5	The CDCN Algorithm	81
5.1	Introduction to the CDCN Algorithm	83
5.2	MMSE Estimator of the Cepstral Vector	86
5.3	ML Estimation of Noise and Spectral Tilt	88
5.4	Implementation Details	90
5.5	Summary of the CDCN Algorithm	93
5.6	Evaluation Results	94
6	Other Algorithms	101
6.1	The ISDCN Algorithm	101
6.2	The BSDCN Algorithm	104
6.3	The FCDCN Algorithm	108
6.4	Environmental Adaptation in Real Time	116
7	Frequency Normalization	121
7.1	The Use of Mel-scale Parameters	121
7.2	Improved Frequency Resolution	123
7.3	Variable Frequency Warping	126
8	Summary of Results	131
9	Conclusions	137
9.1	Contributions	137
9.2	Suggestions for Future Work	139
	Appendix I. Glossary	143
	Appendix II. Signal Processing in Sphinx	145
	Appendix III. The Bilinear Transform	149
	Appendix IV. Spectral Estimation Issues	153
	Appendix V. MMSE Estimation in the CDCN Algorithm	155
	Appendix VI. Maximum Likelihood via the EM Algorithm	161
	Appendix VII. ML Estimation of Noise and Spectral Tilt	165
	Appendix VIII. Vocabulary and Pronunciation Dictionary	169
	References	173
	Index	185