EUROSPEECH 2001 Scandinavia
A novel algorithm for dual channel speech enhancement is presented. It combines the coherence function and a subspace approach in the Bark domain together with an optimal subspace selection through the minimum description length (MDL) criterion. The coherence function allows one to exploit the spatial diversity of the sound field. The processing in the Bark domain permits to take into account of masking properties of the human auditory system while the MDL-based subspace approach ensures statistical robustness. Performance evaluation in real sound fields has highlighted the ability of the algorithm to enhance noisy signals and improve intelligibility for various experimental conditions.
Bibliographic reference. Vetter, Rolf / Renevey, Philippe / Krauss, Jens (2001): "Dual channel speech enhancement using coherence function and MDL-based subspace approach in bark domain", In EUROSPEECH-2001, 1883-1886.