The songs of many birds are unusual in that they serve a role in identifying conspecific mates, yet they are also culturally transmitted. Noting the apparently high rate of diversity in one avian taxon, the songbirds, in which song learning appears ubiquitous, it has often been speculated that cultural transmission may increase the rate of speciation. Here we examine the possibility that song learning affects the rate of allopatric speciation. We construct a population-genetic model of allopatric divergence that explores the evolution of genes that underlie learning preferences (predispositions to learn some songs over others). We compare this with a model in which mating signals are inherited only genetically. Models are constructed for the cases where songs and preferences are affected by the same or different loci, and we analyze them using analytical local stability analysis combined with simulations of drift and directional sexual selection. Under nearly all conditions examined, song divergence occurs more readily in the learning model than in the nonlearning model. This is a result of reduced frequency-dependent selection in the learning models. Cultural evolution causes males with unusual genotypes to tend to learn from the majority of males around them, and thus develop songs compatible with the majority of the females in the population. Unusual genotypes can therefore be masked by learning. Over a wide range of conditions, learning therefore reduces the waiting time for speciation to occur and can be predicted to accelerate the rate of speciation.