An Evaluation Framework for Text-to-Speech Voice Reconstruction

The article presents a new evaluation framework for Text-to-Speech (TTS) voice reconstruction aimed at enhancing communication for individuals with speech disorders while preserving speaker identity. It introduces both subjective and objective assessment methods, including Best Worst Scaling (BWS) for intelligibility and identity, and a novel dual-reference distributional measure to better evaluate the performance of TTS systems. This framework was tested on 17 zero-shot TTS systems across 193 speakers, offering a more reliable and task-aligned approach for practitioners working on voice reconstruction technologies.