Ten subjects participated in the experiment (5 female; mean age  =  23.8 years).

The stimulus set included two conditions: (a) ‘continuous’ stimuli constructed as described in Sect. 2, above, and (b) ‘silent-gap’ stimuli with a 200 ms silent gap inserted at the time of change; gap duration (200 ms) was chosen to be as short as possible so as to minimise reliance on memory capacities but longer than the longest inter-pulse interval (corresponding to the slowest AM rate used) in order to introduce a detectable gap for all scene components. The signals before and after the gap were ramped with a 10 ms cosine-squared ramp. For each gap condition, signals were generated as NC/CD/CA triplets such that NC signals also contained a gap at the same time as their matching CA and CD scenes. Stimulus presentation was blocked by change type and gap type (no gap/silence). The proportion of change events was 50 % in each block. Block order was randomised across listeners. Experimental sessions lasted about 2 h and consisted of a short practice session with feedback, followed by the main experiment with no feedback, divided into runs of about 10 min. Subjects were instructed to fixate at a cross presented on the computer screen and press a keyboard button as soon as possible when they detected a change in the ongoing scene stimulus.

The results are presented in Fig. 21.2. Panel a shows data for continuous scenes – listeners are very sensitive to source appearance but have difficulty detecting disappearing sources (the performance advantage is also reflected in response times; not shown). This finding is likely related to previously reported ‘enhancement’ effects (e.g. Erviti et al. 2011, for review; see also Chap.​ 19) which refer to the finding that local power change, e.g. an increase in the power of one component within a pure-tone chord, results in perceptual pop-out of the associated component away from the rest of the mixture (see also Bregman et al. 1994). Investigations of enhancement phenomena have usually used much simpler stimuli (pure-tone chords) than those used here, and the present demonstration thus suggests that appearance pop-out is a wide-ranging phenomenon, unaffected by the multitudes of onsets and offsets in our scene components.

The CA advantage likely receives contribution from at least two different low-level neural mechanisms: (a) adaptation effects – changes to the sustained neuronal firing rate, which follows the signal while it is present in the scene. Adaptation could thus contribute to onset detection by reducing responses to the ongoing (nonchanging) scene components. (b) Local transients generated at signal onset – auditory onset-/offset-tuned cell populations are characterised by markedly different properties: offset-tuned cells are fewer in number, and their responses tend to be of longer latency and smaller amplitude (Scholl et al. 2010; Phillips et al. 2002) thus leading to larger, and earlier, on- than off-transient responses. These differences are consistent with our finding that appearance events are overall more detectable.