Enhancements in the Multi-Band Excitation (MBE) Scheme for Better Speech Quality at Low Bit Rate Under Noisy Conditions
This paper describes various new algorithms proposed in the MBE scheme for enhancements of speech quality under noisy conditions. The MBE scheme is derived from mainstream sinusoidal coding; where voiced speech is reproduced as a weighted sum of sine waves at the harmonics of the pitch frequency. The MBE scheme represents speech as a product of spectral envelope and the excitation spectrum. The excitation spectrum is represented in terms of parameters such as the fundamental pitch frequency and a voiced/unvoiced decision for each band comprising of a few harmonics in the spectrum. The spectral envelope is the smoothed version of the speech spectrum. The spectral amplitude is encoded based on whether the harmonic is voiced or unvoiced (assuming the excitation to be a sinusoid or narrow band noise for the voiced and unvoiced cases respectively).