Towards a Phonology-Informed Evaluation of Multilingual TTS

Sneha Ray Barman, Neeraj Kumar Sharma, Shakuntala Mahanta
7/2/2026
cs.CLcs.ETcs.LG

No official code found. Best implementation: snehagitrep/TTSEvalVH_interspeech2026 (0 stars, last commit 2026-06-17).

Abstract

Neural TTS systems can sound natural across languages, but naturalness does not guarantee the preservation of sound contrasts that distinguish words from their grammatical forms. Standard metrics like MOS do not test for this. We propose a classifier-based framework that audits TTS output against language-specific phonological patterns using human speech as a benchmark. Testing Assamese advanced tongue root (ATR) vowel harmony with Meta's MMS TTS, we show that a classifier trained on human speech transfers to synthesized speech with minimal loss. The faithfulness audit reveals that [+ATR] mid vowels are realized as [-ATR] in 1/3 tokens despite an underlying [+ATR] specification, a bias absent in human speech. At the word level, predicted ATR labels classify harmony more accurately than transcription labels, indicating a gap between intended and produced phonology. The framework offers task-specific diagnostics and generalizes to other phonological contrasts with measurable acoustic cues.

View on arXivView PDF

Code Implementations(1)

snehagitrep/TTSEvalVH_interspeech2026Community69%

In our paper titled "Towards a Phonology-Informed Evaluation of Multilingual TTS", we propose a pipeline that evaluates Meta's Multilingual TTS to see whether the synthetic outputs contain vowel harmony patterns observed in human speech. Our pipeline uncovers that the studied TTS fails the phonological faithfulness audit.

00Jun 6, 20264 months ago

Similar Papers

Information-Theoretic Characterization of Vowel Harmony: A Cross-Linguistic Study on Word Lists

85% match2023

Multilingual Phonological Feature Recognition with Self-Supervised Speech Models

82% match2026

Deciphering Assamese Vowel Harmony with Featural InfoWaveGAN

81% match2024

ReNikud: Audio-Supervised Hebrew Grapheme-to-Phoneme Conversion

81% match2026

Light-weight Pronunciation Assessment via Discrete Speech Token Surprisal

81% match2026

Phonological Interference in Multilingual Speech Models

81% match2026

Cite this paper

@article{barman2026towards,
  title  = {Towards a Phonology-Informed Evaluation of Multilingual TTS},
  author = {Sneha Ray Barman and Neeraj Kumar Sharma and Shakuntala Mahanta},
  year   = {2026},
  eprint = {2607.01965},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2607.01965v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2607.01965.svg)](https://codeofpaper.com/paper/2607.01965)

Discussion