Pulse signal is an effective indicator to reflect the physiological and physical state of the human body. There are many heart rate estimation methods in videos and most of them manually design algorithm to modeling noise signal, which is not enough to represent the actual distribution of noise. In this paper, we propose to train a two-layer LSTM to estimate pulse signals because long short-term memory (LSTM) can preserve useful signals by filtering out noise signals upon data-driven. In order to overcome the problem of insufficient heart rate public database, we propose to use quantities of synthetic signals which are generated by the algorithm we designed to pre-train the model and pure periodic signals are filtered from LSTM to calculate the heart rate. Experiential results on the public-domain database show the effectiveness of our proposed method that can be a reference for the heart rate estimation.