Audio-Visual Speech Recognition Scheme Based on Wavelets and Random Forests Classification

LUCAS D. TERISSI; GONZALO D. SAD; JUAN CARLOS GÓMEZ; MARIANELA PARODI

LECTURE NOTES IN COMPUTER SCIENCE

Springer

Año: 2015 vol. 9423 p. 567 - 574

0302-9743

This paper describes an audio-visual speech recognition systembased on wavelets and Random Forests. Wavelet multiresolutionanalysis is used to represent in a compact form the sequence of bothacoustic and visual input parameters. Then, recognition is performedusing Random Forests classification using the wavelet-based features asinputs. The efficiency of the proposed speech recognition scheme is evaluatedover two audio-visual databases, considering acoustic noisy conditions.Experimental results show that a good performance is achievedwith the proposed system, outperforming the efficiency of traditionalHidden Markov Model-based approaches. The proposed system has onlyone tuning parameter, however, experimental results also show that thisparameter can be selected within a small range without significantlychanging the recognition results.