A joint speech coding-enhancement algorithm for MBE vocoder

Multi-band excitation vocoder can yield high speech quality at medium and low rate speech coding, but its performance degrades dramatically in low SNR environments. A joint speech coding-enhancement algorithm is proposed which integrates MBE speech coding with speech enhancement. The original MBE coding algorithm is modified for better estimation of the MBE coding parameters in a noisy environment. Especially, the MMSE-LOG speech enhancement algorithm is incorporated in the MBE frequency domain calculation, and speech enhancement using the dual excitation speech model is adopted for post-processing of the synthesized speech. The objective measure and informal listening test indicate that significant improvement can be attained with an input SNR as low as -5 dB