Coherence-based sub-band decomposition for efficient reverberation and noise removal in enclosed sound fields
Joaquín González-Rodríguez, Javier Ortega-García · The Journal of the Acoustical Society of America · 1998
The issue of speech enhancement in reverberant rooms has been extensively addressed. The use of sub-band decomposition through nested linear arrays has been shown to be very efficient for this task. Recently, the coherence/noncoherence nature of the sound field has been used to separate the wideband speech signal in two sub-bands, obtained from the sinc function that characterizes the diffuse field prediction for a given separation between microphones when many independent sources are present and the reverberation time is not too short. The processing scheme proposed here takes advantage of both ideas. After a sub-band decomposition from a linear log-spaced 15 microphone array, a further subdivision is made in each processing band, dividing in a coherent sub-band, where a Generalized–Sidelobe–Canceller is applied, and a noncoherent sub-band, where a Wiener postfilter is used, resulting in a total of six analysis sub-bands. Promising results have being obtained with room simulations (image method), and further results are being obtained with the Carnegie Mellon University (CMU) multichannel-recorded database. The authors thank Dr. Finn Jacobsen (Technical Univ. Denmark) for his fruitful discussions about spatial coherence, and Dr. Richard Stern (CMU, PA, USA) for the multichannel database. [Work supported by CICYT under Project TIC97-1001-C02-01.]