Paper Detail

Beyond Naturalness: Probing Automated Text-To-Speech Evaluators on Linguistically Grounded Dimensions

Oluwanifemi Bamgbose, Simon Rosen, Jash Shah, Lindsay Devon Brin, Hoang H Nguyen, Anke Koelzer, Rachel Hansen, Tara Bogavelli, Fanny Riols

arxiv Score 12.2

Published 2026-08-10 · First seen 2026-08-11

General AI

Abstract

Automated Text-to-Speech (TTS) evaluation methods (Mean Opinion Score (MOS) predictors and Audio Large Language Models (Audio-LLM) judges) are expected to reflect human perception, yet it is unclear how well they capture the distinct aspects of speech that listeners actually perceive. We deconstruct "naturalness" into a linguistically grounded annotation schema spanning 10 distinct perceptual dimensions, and use it to construct the first dimension-level meta-evaluation benchmark for TTS, comprising 860 utterances annotated by trained linguist raters. Results from benchmarking four MOS predictors and four Audio-LLM judges reveal that MOS predictors collapse onto acoustic signal quality, while Audio-LLM judges show selective, prompt-dependent detection that does not generalise across all dimensions. Neither class reliably captures a breadth of linguistically structured speech errors. Our dataset, annotation schema, and evaluation code are publicly released to support more targeted and interpretable TTS evaluation.

Workflow Status

Review status
pending
Role
unreviewed
Read priority
now
Vote
Not set.
Saved
no
Collections
Not filed yet.
Next action
Not filled yet.

Reading Brief

No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.

Why It Surfaced

No ranking explanation is available yet.

Tags

No tags.

BibTeX

@article{bamgbose2026beyond,
  title = {Beyond Naturalness: Probing Automated Text-To-Speech Evaluators on Linguistically Grounded Dimensions},
  author = {Oluwanifemi Bamgbose and Simon Rosen and Jash Shah and Lindsay Devon Brin and Hoang H Nguyen and Anke Koelzer and Rachel Hansen and Tara Bogavelli and Fanny Riols},
  year = {2026},
  abstract = {Automated Text-to-Speech (TTS) evaluation methods (Mean Opinion Score (MOS) predictors and Audio Large Language Models (Audio-LLM) judges) are expected to reflect human perception, yet it is unclear how well they capture the distinct aspects of speech that listeners actually perceive. We deconstruct "naturalness" into a linguistically grounded annotation schema spanning 10 distinct perceptual dimensions, and use it to construct the first dimension-level meta-evaluation benchmark for TTS, compris},
  url = {https://arxiv.org/abs/2608.09930},
  keywords = {cs.SD, cs.AI, cs.CL},
  eprint = {2608.09930},
  archiveprefix = {arXiv},
}

Metadata

{}