A fundamental hypothesis about vocal learning is that young animals learn vocalizations in their natal areas and, following post-natal dispersal, they may introduce new types of vocalizations into their breeding areas. We tested this hypothesis in a tropical bird, the Rufous-and-white Wren (Thryophilus rufalbus), a species in which both sexes produce learned songs. We collected blood samples and acoustic recordings from 146 adult wrens from three populations in northwestern Costa Rica. We genotyped individuals at 10 polymorphic microsatellite loci and identified first-generation migrants using partial Bayesian genotype assignment. We quantified acoustic variation by comparing fine-scale acoustic structure, song sharing, and repertoire novelty between residents and first-generation migrants. We found significant population-level differences in acoustic structure of songs among the three populations. Of the 146 individuals genotyped, 9 individuals were identified as first-generation migrants. In contrast to our predictions, however, we found that these first-generation migrants did not exhibit differences in the acoustic structure of their songs from resident individuals in their breeding population, either for males or females. We conclude that song learning in first-generation migrants must be behaviourly influenced by birds in their breeding populations, following post-natal dispersal. We observed population-level acoustic differences among the three study sites, which implies sustained divergent selection pressures at each site, possibly reflecting acoustic adaptation to different environments or social pressure to sing local songs. Understanding and quantifying patterns of cultural evolution at multiple scales provides insight into how behavioural barriers, such as acoustic signals, contribute to population differentiation and even speciation.