A collaborative speech enhancement approach for speech recognition in motorcycle environment

Aiming at the optimization of the speech recognition performance, we investigate various configurations for a speech front-end, which is part of a multimodal dialogue interaction interface of a wearable solution for information support of the motorcycle police force on the move. Initially, the practical value of various speech enhancement techniques is assessed, and subsequently a collaborative scheme employing independent speech enhancement channels, which operate in parallel on a common input, is proposed. It was experimentally found that the Adaboost.M1 algorithm is the most advantageous among a number of fusion methods. The improvement of speech recognition accuracy due to the collaborative speech enhancement scheme is expressed as gain of 8% in terms of word recognition rate, when compared to the performance of the best speech enhancement channel, alone.