A new vector encoder-decoder (VED) design that is jointly adaptive to source statistics and optimised for noisy channels is proposed. The vector encoder consists of a source-adaptive vector quantiser (AVQ) and a structured index assignment with embedded redundancy (IAER) scheme to provide error-protection capability for noisy channels. The AVQ and IAER processes are combined together resulting in a very effective source-adaptive and channel-optimised VED design. The joint algorithm is designed and evaluated for the non-stationary Wiener process, modelling the source, and for two noisy channels, the binary symmetric channel and the more practical wireless flat-fading Rayleigh channel. It is shown that the proposed technique gives substantial signal-to-noise ratio gains over the separate design of the source AVQ and the IAER in both channel conditions without any additional online computational complexity. In addition, it is suitable for different applications of varying source and channel environments.