TY - GEN
T1 - Adaptive Sparse Linear Prediction in Fixed-Filter ANC Headphone Applications for Multi-Speaker Speech Reduction
AU - Iotov, Yurii
AU - Nørholm, Sidsel Marie
AU - Belyi, Valiantsin
AU - Christensen, Mads Græsbøll
PY - 2023/9/15
Y1 - 2023/9/15
N2 - In some cases, speech can be a disturbing source of ambient noise. Active noise control (ANC) systems have difficulties in dealing with speech due to its non-stationary nature and constraints in the ANC system, which require the optimal filters to be non-causal. The non-causality is due to the delay incurred by, e.g., digital processing or acoustic propagation paths. We propose a fixed-filter feedforward ANC system, HOSpLP-ANC, which aims at attenuating voiced speech in, e.g., office environments. It comprises an adaptive high-order sparse linear predictor (HOSpLP) based on the improved proportionate normalised least mean square algorithm to predict speech ahead in time, thus overcoming such delay. Notably, HOSpLP provides high prediction performance of voiced speech by modelling the joint short- and long-term linear prediction scheme, but without using pitch estimation. This can be of particular significance in the case of the complicated multi-pitch estimation scenario. The results show that HOSpLP-ANC outperforms conventional adaptive feedforward ANC for delays in the order of milliseconds in both single- and multi-speaker environments.
AB - In some cases, speech can be a disturbing source of ambient noise. Active noise control (ANC) systems have difficulties in dealing with speech due to its non-stationary nature and constraints in the ANC system, which require the optimal filters to be non-causal. The non-causality is due to the delay incurred by, e.g., digital processing or acoustic propagation paths. We propose a fixed-filter feedforward ANC system, HOSpLP-ANC, which aims at attenuating voiced speech in, e.g., office environments. It comprises an adaptive high-order sparse linear predictor (HOSpLP) based on the improved proportionate normalised least mean square algorithm to predict speech ahead in time, thus overcoming such delay. Notably, HOSpLP provides high prediction performance of voiced speech by modelling the joint short- and long-term linear prediction scheme, but without using pitch estimation. This can be of particular significance in the case of the complicated multi-pitch estimation scenario. The results show that HOSpLP-ANC outperforms conventional adaptive feedforward ANC for delays in the order of milliseconds in both single- and multi-speaker environments.
KW - Adaptation models
KW - Adaptive systems
KW - Estimation
KW - Headphones
KW - Prediction algorithms
KW - Predictive models
KW - Signal processing algorithms
KW - Speech attenuation
KW - IPNLMS
KW - ANC causality
UR - http://www.scopus.com/inward/record.url?scp=85173031286&partnerID=8YFLogxK
U2 - 10.1109/WASPAA58266.2023.10248065
DO - 10.1109/WASPAA58266.2023.10248065
M3 - Article in proceeding
SN - 979-8-3503-2373-3
T3 - IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA)
BT - Proceedings of the 2023 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, WASPAA 2023
PB - IEEE
T2 - 2023 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA)
Y2 - 22 October 2023 through 25 October 2023
ER -