A uni ed approach to speech enhancement and voice activity detection

In this work, a uni ed system for voice activity detection (VAD) and speech enhancement is proposed. In the proposed system, there is mutual exchange of information between VAD and speech enhancement blocks. A new, robust and low complexity VAD algorithm is implemented for the VAD block of the uni ed system. The newly proposed VAD algorithm uses a periodicity measure and an energy measure obtained from spectral energy distribution and spectral energy di erence of the input speech data. For the speech enhancement block, the Modi ed Wiener Filtering (MWF) algorithm is utilized. It has been shown that the utilization of information exchange between the VAD and MWF algorithms in the uni ed system increases the performance of both algorithms and the proposed uni ed system improves the robustness of a speech recognition system signi cantly. Both of the enhanced algorithms are non-iterative. Therefore, the proposed uni ed system is computationally attractive for real-time applications.

Autorentext

Ceyhan Kasap received his B.Sc. (with high honors) and M.Sc. degrees in Electrical and Electronics Engineering from Bogazici University - Istanbul, Turkey, in 2007 and 2009 respectively. His main research interests are speech processing and algorithm design. He has been working in the software development industry since 2007.

Klappentext

In this work, a uni ed system for voice activity detection (VAD) and speech enhancement is proposed. In the proposed system, there is mutual exchange of information between VAD and speech enhancement blocks. A new, robust and low complexity VAD algorithm is implemented for the VAD block of the uni ed system. The newly proposed VAD algorithm uses a periodicity measure and an energy measure obtained from spectral energy distribution and spectral energy di erence of the input speech data. For the speech enhancement block, the Modi ed Wiener Filtering (MWF) algorithm is utilized. It has been shown that the utilization of information exchange between the VAD and MWF algorithms in the uni ed system increases the performance of both algorithms and the proposed uni ed system improves the robustness of a speech recognition system signi cantly. Both of the enhanced algorithms are non-iterative. Therefore, the proposed uni ed system is computationally attractive for real-time applications.