INVESTIGADORES
DI PERSIA Leandro Ezequiel
congresos y reuniones científicas
Título:
Pseudoanechoic blind source separation with improved Wiener postfilter
Autor/es:
LEANDRO EZEQUIEL DI PERSIA; DIEGO HUMBERTO MILON; MASUZO YANAGIDA
Lugar:
Rosario
Reunión:
Congreso; RPIC 2009 XIII Reunión de trabajo en Procesamiento de la Información y Control; 2009
Institución organizadora:
Laboratorio de Sistemas Dinámicos y Procesamiento de la Información, FCEIA, UNR
Resumen:
The pseudoanechoic model wasproposed recently to simplify the parameterestimation in blind source separation basedon frequency-domain independent componentanalysis. In the method, after separation basedin the pseudoanechoic model a time-frequencyWiener postfilter to improve the separation isapplied. In this contribution, a deeper analysisof the working principles of the Wiener postfilteris presented. Furthermore, a variationof this postfilter to improve the performanceusing the information of previous frames is introduced.The improvements obtained throughthe new method are evaluated in an automaticspeech recognition task and with the PESQ objectivequality measure. The results show anincreased robustness and stability of the proposedmethod.
