The receive QMFs shown in G.722 are two linear-phase non-recursive digital filters which interpolate the outputs, [r.sub.L] and [r.sub.H], of the lower and higher sub-band ADPCM
decoders from 8 kHz to 16 kHz and which then produce an output, [x.sub.out], sampled at 16 kHz which forms the input to the receive audio parts.
where a and b are estimated by the least squares fit of the distortion D and PESQ pairs of ADPCM
Among them, adpcm
and MatMul show better scalability, while some other benchmarks show worse, such as backprop or 183.equake.
The greatest reduction in size is about 40% for the adpcm
program, while the smallest is about 15-17% for the go program.
While DPCM usually gives at least 2:1, ADPCM
gives higher reductions.
The standard speech coder starts from the 64kbps PCM (Pulse Code Modulation) method adopted in ITU-T recommendation G.711 in 1972 and standardized to 32kbps ADPCM
(Adaptive DPCM) and 16kbps LD-CELP (Low-Delay CELP).
A recommended definition of the 32 kb/s Adaptive Differential Pulse Code Modulation (ADPCM
) algorithm was published by International Telephone and Telegraph Consultative Committee (CCITT, the recent name is International Telecommunication Union, ITU) as recommendation G.721 (CCITT Recommendation G.721, 1984).
The benchmarks are jpeg, mpeg, gzip, susan, gsm, adpcm
, and blowfish programs.
If the degree of correlation between consecutive samples varies with time, then ADPCM
can be used, where adaptation of the predictor coefficients is done according to the correlation between samples , .
Access to software libraries is also available for added functionality with speech compression for protocols such as G.711 (64 kbps), ADPCM
G.726A (16-40 kbps) and SPEEX (8 kbps).