This paper listens to the deaf machine—and to those who sing before it. In South Korean noraebang (karaoke rooms), the transmediated voice is scored by rating systems whose quantitative assessments bear no systematic relation to vocal performance—scores are generated randomly, unconnected to any measurable quality of the sung voice. Yet despite widespread awareness of this arbitrariness, scores continue to orient performance, inciting competitive transformations as singers modify their vocal delivery to appease what they believe the algorithms of deaf machines can discern. Drawing from fieldwork in Seoul, I examine how the persistent sociality of random scoring extends the pervasive audit culture (Shore & Wright 2024, Strathern 2000) that infiltrates Korean subject-formation through compounding regimes of examination, ranking, and spec accumulation (Cho 2015). As the neoliberalization of language reduces communicative faculties to assessable skills (Park 2021, Urciuoli 2008), and semiocapitalism abstracts language from affective bodies (Berardi 2018), I trace how even leisure sociability—where the voice might otherwise exceed productive functionalization—becomes saturated with the competitive logics of quantitative comparison. Singers perform simultaneously for the machine and for the room, compressing the poetic intensities of sung language into discrete, machine-legible aesthetics that double as competitive display. Such logics go largely unremarked in the ordinary sociality of noraebang. Yet when my interlocutor, Seohyeon, disabled the system to sing without numerical subjection, her quiet refusal (Campt 2017) of the quantitative assessment momentarily reactivated the affective excess of poetic language and the communal intimacies of song otherwise obscured by Korea’s broad culture of audit and comparison.