Orthogonal time frequency space (OTFS) modulation had a lower bit error rate than orthogonal frequency division multiplexing (OFDM) in delay-Doppler (DD) channels. In order to obtain OTFS modulated data, channel parameters needed to be estimated. The commonly used method for channel estimation was to embed pilot symbols between information symbols and added sufficient protection intervals along the delay and Doppler axis. However, this method was often difficult to accurately estimate the unavoidable fractional Doppler frequency offset in practice, resulting in an increase in error rate. This paper combined pilot symbols with MUSIC algorithm and proposed an embedded pilot-multiple signal classification (EPMUSIC) algorithm. This algorithm can accurately estimate fractional Doppler while avoiding the significant computational complexity brought by the MUSIC algorithm during peak search. Through the simulation, the performance of the proposed algorithm was better than that of the traditional pilot algorithm, slightly inferior to the MUSIC algorithm, but the computational complexity is greatly reduced.