To estimate the heart rate (HR) from electrocardiogram (ECG), time-frequency representations such as the short-time Fourier transform (STFT) is often used. As the STFT is constrained by the choice of a specific analysis window used for its definition, we alternatively propose to estimate HR from a synchrosqueezed STFT. More precisely, we build a novel algorithm inspired by non-negative matrix factorization to estimate HR by determining the minimal Wasserstein distance between a synchrosqueezed STFT and columns of a specific dictionary matrix. Throughout numerical simulations carried out on both synthetic and real ECGs, we show in what way to use a synchrosqueezed STFT rather than STFT improves HR estimation.