This thesis radically breaks away from previous ideas on the voice and ethics in documentary (Nichols 2001, Rangan 2017) and proposes a new framework for analysing the non-linguistic voice in interview-based non-fiction film. This new framework expands on and applies Emmanuel Levinas’ concept of saying </em>(1976, 1979, 1981), and the relational theory of Adriana Cavarero (2005, 2014, 2021) and Judith Butler (1988, 1991, 1997, 2021) to the production of the voice in the face-to-face encounter between interviewer and interviewee. The proposed relational mode of analysis asks listeners to listen otherwise to the audible breaths, sighs, arr’s and stammers, which the thesis claims initiate a nuanced and in-depth reading of interview encounters that acknowledge the intimacy, collectivity and uniqueness of each voice within the interview exchange. This approach to documentary analysis through relational theory forges the connection between how </em>voices are produced in practice during the face-to-face encounter with what is heard in the finished film. Through a close reading of the documentaries: About a War </em>(Rugo and Weaver, 2018); Chronicle of a Summer </em>(Morin and Rouch, 1961); Titicut Follies </em>(Wiseman, 1967); Shoah </em>(Lanzmann, 1987); Duch, Master of the Forges of Hell </em>(Panh, 2011); S-21: The Khmer Rouge Killing Machine</em> (Panh, 2003); Survive: In the Heart of the Khmer Rouge Madness</em> (Saidnattar 2009); The Fog of War </em>(Morris 2003) and <em>Aileen Wornous: The Life and Death of a Serial Killer</em> (Broomfield and Churchill, 2004) the thesis develops terms such as polyphony, collectivity, shared space, failed confession, face-to-face, saying, </em>proximity, vocal agency and methodology with particular focus on how perpetrator’s voices in documentary films de-centre existing theory on ethics and the voice in the documentary filmed interview.