Sound specific modelling and synthesis with a new postfiltering in low bit rate speech coding
Rodrigo C. de Lamare, L.M. da Silva, Abraham Alcaim · 2003
In this paper we investigate the use of fricatives and stops modelling and synthesis techniques with a spectral envelope reconstruction combined with noise reduction postfilter (SERNR) in mixed voiced-unvoiced multiband excitation coders. We perform a comparative analysis amongst a noise excitation approach operating at 1.75 kb/s and a fricatives and stops excitation technique operating at 0.4 kb/s. A novel SERNR postfiltering technique that significantly enhances the decoded speech is proposed and compared with the well-known adaptive spectral enhancement (ASE) filter.