Blocked-cyclic picture naming is a prominent tool in word production research. It was found that responses were slower when the stimuli were presented blocked by semantic category than when they are presented intermixed. Wöhner et al. (2021) extended the task to sound naming and found that the semantic interference was much larger in sound naming (e.g., a barking) than in picture naming (e.g., of a dog), by a factor of approximately three, even when the same words were produced (e.g., “dog”). We explored possible reasons for this differential interference pattern. We replicated the differential interference when the stimuli were presented briefly, ruling out that it results from substantial overlap between stimulus processing on the one hand and phonological response preparation and pre-articulatory self-monitoring on the other. In contrast, we found similar-sized interference when the sounds and pictures were replaced with spoken and written words to which participants responded with the translation in their second language, ruling out that the differential interference results from different input modalities. We also observed the differential interference in sound and picture naming when the spoken response was replaced with a typed response, ruling out that it results from a specific (spoken) response modality. Overall, the data demonstrate the robustness of the differential interference in sound and picture naming and provide evidence that the semantic interference in blocked-cyclic naming is effectively shaped by early stimulus-specific perceptual and semantic – “lead-in” – processes, which precede lexical selection. These early processes require more attention in word production research.