This paper is aimed to the study of spontaneous radiant energy emitted by trapped vortex flame, focusing on the intermittent occurrence of flame blowout during thermoacoustic oscillation. We tackle this issue by a wavelet–based auto–conditioning technique, able to detect intermittent structures in the radiant energy signal, which we identify as a trace of extinction–reignition events. Intermittent events are found to be in phase with the tonal component of the radiant energy signal, although they occur randomly in time. Despite the wavelet auto–conditioning technique is a well established method in fluid dynamics and aeroacoustics, our work represents the first application in combustion field.Additional information about blow–out are inferred by analysing the average signature computed from the radiant power time series: re–ignition velocity appears higher than extinction velocity. This result confirms the effectiveness of the radiant power for combustion diagnostics.