We study the downlink spectral efficiency (SE) of a cell-free massive multiple-input multiple-output system with orthogonal time frequency space (OTFS) modulation. Each user equipment (UE) uses the minimum mean-squared error-based successive interference cancellation scheme to detect the received signal. Given the statistical channel information at the UEs, an analytical closed-form expression for the downlink SE is derived. To improve the per-user SE, a max-min fairness resource allocation is proposed and developed. Numerical results show that the proposed resource allocation strategies for superimposed pilot-based and embedded pilot-based channel estimation achieve up to 43% and 65% performance gain over the baseline system without power allocation.