Languages change in multiple ways, from syntax and lexicon to prosody and local phonetic features. The indexicality of those forms depends both on the given social context and what other forms are present. With multiple sources of variation, studying sociolinguistic change is challenging. Instead of analyzing variation of specific phonetic features in a small set of ‘representative’ speakers, this paper presents a purely speech signal-based analysis of Finnish dialects and how gender and age contribute to dialectal variation by analyzing the latent embedding space of a large self-supervised model fine-tuned for language identification. Our results show that the embedding space contains meaningful relationships on the dialect level and suggest that males tend to exhibit clearer dialectal traits than females, with young women on one end of the continuum and older men on the other. The study shows promise for quantifying sociolinguistic change with data from the ‘wild’ on a large scale, complementing the established methodologies.