Sampling schemes on the sphere require O(L^2) samples to perfectly sample and reconstruct a signal with bandwidth L. If the signal in question is a low-pass observation of a finite collection of spikes or rotations of a known function, we can use less samples. We propose an algorithm that improves over the best known finite rate of innovation (FRI) sampling scheme on the sphere by a factor of approximately four. Further, we show how multiple sound source localization (SSL) by a spherical microphone array can be transformed into a spherical FRI sampling problem. We certify the effectiveness of the proposed algorithm by using it to solve the SSL problem.