Within the emerging new technologies in Smart Grids, the concern about Non-Intrusive Load Monitoring (NILM) techniques has increased over the last five years. These techniques aim to disaggregate the energy consumption on a household or commercial building into individual appliances from the use of a single-point sensor. Most efforts attempt to disaggregate by using a low sampling frequency and by focusing on steady consumption states. Nevertheless, to achieve a better accuracy, it is necessary a higher sampling frequency and to look up both steady states and transient states. This paper proposes a novel event-based detector that could be used in NILM systems in a previous stage in order to detect transient states and help to the labelling task. This could lead to a better performance, specially taking into account small power changes which are not usually well detected on most low-frequency systems. It is based on the use of Hilbert Transform to obtain the envelope of the signal and the use of an average and a derivate filter to obtain a set of spikes that characterize each transition. The performance has been tested using a set of known metrics obtaining good results in two cases: sampling at 1kHz, which the aim of a possible integration in smart meters; and in a real scenario where 20 different appliance-loads are connected to the power main.