Although filled pauses (FPs) have been argued to be an unconscious and highly idiosyncratic feature (Boer; Heeren 2019), recent sociolinguistic works show that FPs can be operationalized as a sociolinguistic variable (Fruehwald 2016), exhibit regular social stratification (e.g., Wieling et al 2016), are acoustically different across languages and can be acquired in language contact settings (e.g., Erker; Bruso 2017). This paper investigates the realization of non-lexical FPs in a dialect contact situation in Brazilian Portuguese, as the mid-open fronted vowel [ɛ] vs. the more central vowel [a, ɐ, ɐ̃], analyzed as categorical (eh vs ah) and gradient variables–vowel height (F1), vowel frontness (F2), and vowel duration (ms). The main objectives are to determine whether there are dialectal differences in Brazilian Portuguese regarding FPs; and if so, whether Northeastern migrants living in São Paulo accommodate to the Paulista variant and which social factors predict acquisition. Assuming FPs are below the level of consciousness, this paper also discusses the role of salience in dialect contact (Trudgill 1986, Nycz 2018, Erker 2022). FPs were thus analyzed in a corpus of 40 Northeastern migrants living in São Paulo, balanced for gender, age of arrival, and length of residence, in contrast with 24 non-mobile natives (12 Northeasterners-NE and 12 Paulistas-SP). Analyses of 1,137 tokens of FPs show that eh is the preferred variant in all samples, but with significant differences for the rates of ah (NE: 2.3%; migrants: 7.8%; SP: 18.2%; χ2 = 37.6(2), p<0.001). Acoustic analyses of FPs vowel formants, contrasted with 224 tokens of stressed lexical [ɛ] (e.g., café ‘coffee’, used as control), extracted from Praat (Riebold 2013) and Lobanov-normalized, show that, compared to the respective control-[ɛ], non-migrant Northeasterners’ FPs are higher and more fronted; Paulistas’ FPs are more central; and migrants’ FPs are acoustically similar (Figure 1). No difference in vowel duration was detected. The migrants’ higher rates of ah and more central FPs compared to non-migrant Northeasterners signal convergence to the host community’s FPs. Mixed-effects models of the four response variables in the Paulista data (Table 1), including speakers’ gender, age, and social class as fixed effects and speaker as a random effect, indicate that FPs are a stable feature in the host community. Upper-middle class speakers favor ah and produce higher FPs, and males produce less fronted vowels. In the migrants’ analyses of the effect of gender, age, social class, age of arrival, and length of residence (Table 2), upper-middle class speakers favor ah, less fronted and longer vowels; and males favor less fronted and shorter FPs. There were no social correlations with vowel height and, interestingly, there were no correlations with migrants’ age of arrival or length of residence, differently from previous studies of phonetic variables (e.g., Chambers 1992). The analyses thus show that a low salience and relatively infrequent feature can be acquired in dialect contact, and suggest, from the similar correlations with gender and social class in the host and migrant datasets, that acquisition depends on the migrants’ social networks in São Paulo.