MAD Speech: Measures of Acoustic Diversity of Speech
Matthieu Futeral, Andrea Agostinelli, Marco Tagliasacchi, Neil Zeghidour, Eugene Kharitonov
TL;DR
MAD Speech introduces a modular, per-facet approach to quantify acoustic diversity in generated speech across voice, gender, emotion, accent, and background noise. By mapping utterances to generic embeddings and applying lightweight per-facet projection heads, the method enables facet-specific diversity metrics computed via mean pairwise dissimilarity or Vendi Score. The authors validate the approach with ground-truth datasets and show that MAD Speech correlates more strongly with true diversity than baselines, while revealing how model choices and sampling affect diversity. The framework provides a practical tool for evaluating and guiding generative speech systems and is publicly released for broader use.
Abstract
Generative spoken language models produce speech in a wide range of voices, prosody, and recording conditions, seemingly approaching the diversity of natural speech. However, the extent to which generated speech is acoustically diverse remains unclear due to a lack of appropriate metrics. We address this gap by developing lightweight metrics of acoustic diversity, which we collectively refer to as MAD Speech. We focus on measuring five facets of acoustic diversity: voice, gender, emotion, accent, and background noise. We construct the metrics as a composition of specialized, per-facet embedding models and an aggregation function that measures diversity within the embedding space. Next, we build a series of datasets with a priori known diversity preferences for each facet. Using these datasets, we demonstrate that our proposed metrics achieve a stronger agreement with the ground-truth diversity than baselines. Finally, we showcase the applicability of our proposed metrics across several real-life evaluation scenarios. MAD Speech is made publicly accessible.