Modern developments in speech coding

Modern speech coding algorithms are based on models which attempt to successfully describe short-term speech signals. These models can be defined in the `time´ or in the `frequency´ domain and analysis by synthesis (AbS) estimation procedures are often used to determine the short term model parameters. The parameters can then be vector quantised if the coder is designed to operate at low bit rates (less than 8 kbits/sec). Frequency domain algorithms decompose the speech signal into sinusoidal components, with time varying amplitudes and frequencies, and hence the speech is modelled as a time varying line spectrum. Medium to low bit rate frequency domain coding systems include: adaptive transform coding, subband coding, harmonic coding, sinusoidal transform, multiple excitation, codebook excitation, backward excitation recovery