<p>An audio coder is arranged to process a respective set of sampled signal values for each of a plurality of sequential segments of an audio signal (x). The coder comprises an analyser (TSA) arranged to analyse the sampled signal values to provide one or more sinusoidal codes (Cs) corresponding to respective sinusoidal components of the audio signal. A subtractor subtracts a signal corresponding to the sinusoidal components from the audio signal to provide a first residual signal (r1). A modeller (SEG) models the frequency spectrum of the first residual signal (r1) by determining first filter parameters (Ps) of a filter which has a frequency response approximating a frequency spectrum of the first residual signal. Another subtractor subtracts a signal corresponding to the first filter parameters from the first residual signal to provide a second residual signal (r2). Another modeller (RPE) models a component (r2,r3) of the second residual signal with a pulse train coder (RPE) to provide respective pulse train parameters (L0). A bit stream generator (15) generates an encoded audio stream (AS) including the sinusoidal codes (Cs), the first filter parameters (Ps) and the pulse train parameters (L0).</p>