ESSV Konferenz Elektronische Sprachsignalverarbeitung

Title: Evaluation of single- and dual-channel noise power spectral density estimation algorithms for mobile phones

Authors: Christian Herglotz, Marco Jeub, Christoph Nelke, Christophe Beaugeant, Peter Vary


Noise suppression has become a standard option for high end mobile phones. One essential component for most of these algorithms is the estimation of the noise power spectral density (PSD) of the unwanted background noise. A large number of single and dual-channel approaches that exploit a large variety of different signal properties have been presented in literature. So far, a comprehensive evaluation of these approaches in realistic and reproducible situations does not exist. In this contribution, the performance of well-known as well as recently proposed noise PSD estimators is analyzed with respect to an integration in dual-microphone mobile phones. Additionally, a new algorithm is proposed which exploits explicitly the dual-microphone configuration of state-of-the-art mobile phones and smartphones.

Year: 2011
In session: Sprachsignalverarbeitung, Spracherkennung und Sprachsynthese I
Pages: 13 to 22