A computationally efficient noise estimation algorithm for speech enhancement

Abstract

We introduce a computationally efficient noise estimation algorithm for speech enhancement. The algorithm estimates the background noise during the pauses of the speech signal. However, unlike conventional voice activity detectors, the computational complexity and hardware requirement of the proposed algorithm are very low. In this algorithm, the ratio between the total energy and zero crossing rate over a block of data is used to detect the pauses. The pauses are detected after noise reduction and hence the performance is improved.