An enhanced multiband excitation speech coder at 2,400 b/s
Keith A. Teague · 2003
The design and implementation of a 2400 b/s enhanced multiband excitation (EMBE) speech coder is described. The coder uses a variation of the multiband excitation (MBE) model originally proposed by Griffin and Lim to produce natural sounding and intelligible speech. A pitch-adaptive variable band structure is used for representing voicing decisions, with perceptually weighted spectral smoothing of harmonic amplitudes, and efficient vector quantization of speech spectrum using a four-way split vector quantizer to achieve high quality speech at 2,400 b/s. Objective performance results, in the form of DAM and DRT scores, are presented for quiet and office environments, indicating this coder is well suited for applications requiring low rate communications quality speech.