A speech coding technique has been developed to provide a method of digitising speech at bit rates in the region of 8000 bit/s which is insensitive to the effects of acoustic background noise and bit errors on the digital link. The main aim of the work has been to develop a coding scheme which provides speech quality and robustness against noise and errors that is similar to a 16 000 bit/s CVSD coder, but which operates at half its data rate. The result is a sub-band coder (SBC), with seven bands covering the frequency range 330 to 3200 Hz. The design is unusual in that it does not employ quadrature mirror filters (QMFs) but instead uses parallel finite-impulse-response (FIR) filters which are implemented efficiently with discrete Fourier transforms (DFTs) and prefiltering or windowing which is common to a group of filters. These DFT-based filters naturally provide output samples which are complex baseband vectors, and the quantisers are designed to handle these vectors in an optimum way that would not be possible with the purely real signals from a conventional QMF. The phase and amplitude components of the vectors are coded in ways that attempt to minimise the subjective effects of quantisation distortion and of bit errors between the encoder and decoder. Informal subjective tests indicate that the speech is of good ‘communications’ quality, and the coder can tolerate high levels of acoustic noise and a random bit error rate up to 10%.