Distant talker speech commands recognition accuracy degrades significantly when the speech is corrupted by background sound (music, radio, TV sound, etc). The purpose of this paper is to compare the effect of different noise reduction algorithms on the performance of the recognition system. To achieve high recognition performance for a wide variety of noise types and various signal to- noise ratios, this paper presents two “echo reduction” stereo algorithms: Time Domain Adaptive Filter (TDAF) and Frequency Domain Adaptive Filter (FDAF). New modification of FDAF is proposed. We have evaluated this approach first on computer generated anechoic mixtures and then on real echoic mixtures recorded in a room. FDAF shows much more positive effect. By processing noise-corrupted commands in this manner we achieve significant improvements in spoken commands recognition accuracy. The resulting level of errors (WER) is about 10 % when signal-to-interference ratio is about 0 dB.
Cite as: Koval, S.L., Stolbov, M.B., Tatarnikova, M.Y. (2004) Integration of adaptive noise cancellation for isolated word recognition in smart-home control systems. Proc. 9th Conference on Speech and Computer (SPECOM 2004), 108-111
@inproceedings{koval04_specom,
author={S. L. Koval and M. B. Stolbov and M. Y. Tatarnikova},
title={{Integration of adaptive noise cancellation for isolated word recognition in smart-home control systems}},
year=2004,
booktitle={Proc. 9th Conference on Speech and Computer (SPECOM 2004)},
pages={108--111}
}