Abstract

This article presents ethnographic research in the field of algorithmic speech synthesis, conducted in universities in Italy and Germany and in a speech synthesis company. The methodology integrates ethnography, software studies and media archaeology in order to account for the complex socio-technical network of algorithmic systems. Fieldwork included interviews with programmers as well as the examination of speech synthesis algorithms at work. Following analysis of the empirical research, the study discusses epistemological and socio-cultural aspects of data assemblages, focusing on changes in programming practices related to deep learning, a technology that bypasses domain-knowledge and human models of speech to refer directly to the observation of examples. Highlighting the tension between technical operations and social representations of these operations, the paper suggests that the sense-making of algorithms is not to be found in automation, but in the shift in programmers’ position and in the associated subjectivation processes.