This dataset comprises simulated extracellular spiking neural signals, for which the activity of the active neurons is known. These data can thus be used to test spike sorting algorithms. Each file corresponds to one signal with its ground truth, all stored as a Matlab format file .mat. Each file contains 2 Matlab structures, one for the signal and one for the ground truth (struct signal and struct truth). Each signal lasts 200 seconds with a simulated sampling frequency of 20kHz. Three action potential waveforms are embedded in each signal, two with a maximum amplitude of 5 and one with a maximum amplitude of 10. The dataset is composed of 3 different sets of signals. The first set is made of 70 signals with 7 different noise levels (between 0.5 and 2) and 10 signals for each noise level. In all these signals, each neuron fires at a mean rate of 3.3Hz. The second set is made of one signal with the 3 neurons having different firing rates (1 Hz, 3 Hz, and 9 Hz) The third set is made of one signal with the 3 neurons having the same firing rate of 3 Hz but with intermittent activity.