Speech enhancement using new iterative minimum statistics approach

In hands-free mobile communication, speech quality is often degraded due to presence of surrounding noise. This paper introduces an improved version of Minimum Mean Square Error (MMSE) noise estimator. Noise spectrum estimation is a crucial element used in speech recognition systems. Our proposed noise estimation method is based on a popular searching algorithm used in software engineering called Binary Search (BS), which we integrate with First-In First-Out (FIFO) MMSE noise reduction algorithm. In the literature, there is no research addressing the integration of BS algorithm with an active noise cancellation system. The noise spectral minima are computed using BS algorithm which makes it fast and efficient. The proposed algorithm is tested using real time data collected from vehicles running at different speeds. Simulation results are provided, and it is shown that the proposed algorithm outperforms other MMSE algorithms.