A comparative study of onset detection methods in the presence of background noise

The paper presents a comparative study of five onset detection methods. The comparison was performed in the adverse acoustic conditions where the source signals were mixed with four different background noises. For two source signals: a set of isolated words and a sequence of tones, the influence of background noise at different SNR levels has been measured. Additionally, an experiment with auditory filter bank which comprises gammatone filters has been performed to determine the influence of the noise components on the detection efficacy in various bands. The obtained results show that the final onset detection accuracy can be improved by the fusion of onsets from different gammatone channels.