Controle Ativo de Ruído com Baixo Custo Computacional em Headsets com Reprodução de Fala

This work evaluates the performance of the adaptive algorithms LMS, NLMS and two implementations of APA for active noise control (ANC) in headsets with speech reproduction and contamination by white and babble noises. Besides the mean square error, the algorithms are evaluated by the quality and intelligibility of the processed speech signals. Simulations are performed in time-varying scenarios, where acoustic paths are changed during operation. Results show that the algorithms provide equivalent white noise attenuation, close to 10.7 dB when SNR is 0 dB, with slight superiority of APAs in quality and intelligibility at low SNRs. For babble noise, the APAs stand out for attenuating more than 16 dB when SNR is 0 dB. It is concluded that these low-cost algorithms can improve speech intelligibility at low SNRs, a crucial factor for headsets designed for communication services such as telemarketing and call centers, and quality at high SNRs.