Algorithms for Near-end Listening Enhancement (NELE) improve the intelligibility of speech, played back in a noisy environment, by adaptively filtering the speech signal and taking into account the current background noise characteristics. This contribution proposes a time-domain approach for NELE which is based on linear prediction (LP). LP coefficients of speech and noise are estimated by a normalized least mean squares algorithm and are used to adapt time-domain filters which enhance the speech signal. The computational cost of this approach is comparable to reference algorithms in the frequency domain, but its latency is extremely low. Simulations show that significant improvements in terms of the Speech Intelligibility Index can be achieved.
展开▼
