Serial recall from working memory is known to be impaired by the presence of irrelevant background speech, but several prior studies have concluded that the magnitude of the impairment is independent of the phonological relationship between to-be-remembered (TBR) and to-be-ignored (TBI) sources of information. In the present study, we examined the influence of between-stream phonological similarity in serial recall while attending to a heretofore uncontrolled variable, the phonetic feature. We found that TBI items sharing many phonetic features with TBR items produced significantly stronger working-memory impairments than TBI items with minimal phonetic feature overlap. In addition, participants were more likely to report remembering incorrect items that incorporated phonological characteristics of the TBI stream in the high-overlap condition. These findings provide evidence for subphonemic between-stream interactions and suggest that multiple parallel processes contribute to the irrelevant speech effect. We propose that a 2-component model, which combines the assumptions of process- and content-based accounts for the irrelevant speech effect, offers the best explanation for these findings.