Electromyographic signals (EMG) are widely used in Human Machine Interface applications, in the control of myoelectric prostheses and in the control of models in virtual reality environments. To do so, it is necessary to process the EMG signals in order to extract the necessary information for each application. One of the key steps in the processing of EMG signals is the accurate detection of the beginning and end of muscle contractions. Thus, it is necessary to use methods and algorithms that aim to accurately detect onset and offset times of muscle activity, using techniques that involve the calculation of the signal envelope, calculation of thresholds and digital filters. The present work aims at the development of a system capable of performing the acquisition and plotting of EMG signals, as well as onset and offset detection of muscle activity. The system is consisted of hardware, in order to acquire the signal, and software, which is based on parallel processing for real-time detection. The processing method proposed consists in the application of the Hilbert transform with a low-pass filter to calculate the envelope. There was tested two approaches for the smoothing of the rectified signal, being the moving average algorithm the one which showed better results. The methods used in this work present satisfactory results even in a computer with lower power processing, besides it was developed in a reusable way, allowing the interaction with other software applications.