A human-centered framework for multilingual AI dubbing : quality metrics, taxonomy, and evaluation checklists

Abstract

This paper proposes a human-centered, multidimensional framework to support practitioners and trainees in producing and evaluating multilingual AI dubbing. It introduces a systematic suite of modular tools for professional localization workflows and academic training. This toolkit expands the Script, Speech, and Sound (SSS) Quality framework (Spiteri Miggiani, 2024) by integrating three new measures: a Vocal Identity Taxonomy (VIT) to systematically map vocal persona and voice casting; a Synthesis Validation Checklist (SVC) to identify and mitigate synthetic audio artifacts and processing glitches; and a holistic product-level Express QC Model utilizing “Approve, Improve, Redo” metrics. The paper outlines practical strategies to implement this extended AI dubbing quality framework within industry quality control pipelines. Furthermore, it demonstrates how emerging hybrid workflows redefine the linguist's role, positioning the human expert as an indispensable decision-maker in navigating linguistic, emotional, and cultural resonance. To facilitate practical application and provide conceptual foundations for this evolving field, the article includes supplementary materials, comprising practical templates, an operational toolkit, and a glossary of key terms.