The coder described uses a full-duplex, five-octave-band structure. An additional splitter is used to discard the top half of the highest band, allowing significant data reduction with little loss in quality. Conjugate quadrature filters (CQFs) were chosen for this implementation because they exhibit perfect reconstruction and were expected to offer better performance than simple quadrature mirror filters (QMFs). Adaptive pulse-code modulation (APCM) coders are used to quantize and code each band. The number of levels in the APCM coder is a parameter of the implementation, so the realization can be easily configured to operate at data rates from 10 kb/s to 24 kb/s. Results indicate that the speech quality of the CQF implementation was barely discernible from that of a QMF-based implementation. The basic building block used in the subband coder is the two-band analysis/reconstruction system. The analysis section uses high and low pass filters to split the speech into two bands. Each subband signal is then decimated to its nominal Nyquist rate, thus preserving the total number of samples in the signals to be coded. The reconstruction section inserts zeros into the sample streams and interpolates them using the corresponding reconstruction filters. >