Guardado en:
Detalles Bibliográficos
Autores principales: Shen, Shengfan, Wu, Di, Song, Xingchen, Zhou, Dinghao, Xue, Liumeng, Meng, Meng, Luan, Jian, Wang, Shuai
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:https://arxiv.org/abs/2603.24430
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866915891034193920
author Shen, Shengfan
Wu, Di
Song, Xingchen
Zhou, Dinghao
Xue, Liumeng
Meng, Meng
Luan, Jian
Wang, Shuai
author_facet Shen, Shengfan
Wu, Di
Song, Xingchen
Zhou, Dinghao
Xue, Liumeng
Meng, Meng
Luan, Jian
Wang, Shuai
contents Reliable evaluation of modern zero-shot text-to-speech (TTS) models remains challenging. Subjective tests are costly and hard to reproduce, while objective metrics often saturate, failing to distinguish SOTA systems. To address this, we propose Iterate to Differentiate (I2D), an evaluation framework that recursively synthesizes speech using the model's own outputs as references. Higher-quality models exhibit greater resilience to the distributional shift induced by iterative synthesis, resulting in slower performance degradation. I2D exploits this differential degradation to amplify performance gaps and reveal robustness. By aggregating objective metrics across iterations, I2D improves discriminability and alignment with human judgments, increasing system-level SRCC from 0.118 to 0.464 for UTMOSv2. Experiments on 11 models across Chinese, English, and emotion datasets demonstrate that I2D enables more reliable automated evaluation for zero-shot TTS.
format Preprint
id arxiv_https___arxiv_org_abs_2603_24430
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Iterate to Differentiate: Enhancing Discriminability and Reliability in Zero-Shot TTS Evaluation
Shen, Shengfan
Wu, Di
Song, Xingchen
Zhou, Dinghao
Xue, Liumeng
Meng, Meng
Luan, Jian
Wang, Shuai
Sound
Reliable evaluation of modern zero-shot text-to-speech (TTS) models remains challenging. Subjective tests are costly and hard to reproduce, while objective metrics often saturate, failing to distinguish SOTA systems. To address this, we propose Iterate to Differentiate (I2D), an evaluation framework that recursively synthesizes speech using the model's own outputs as references. Higher-quality models exhibit greater resilience to the distributional shift induced by iterative synthesis, resulting in slower performance degradation. I2D exploits this differential degradation to amplify performance gaps and reveal robustness. By aggregating objective metrics across iterations, I2D improves discriminability and alignment with human judgments, increasing system-level SRCC from 0.118 to 0.464 for UTMOSv2. Experiments on 11 models across Chinese, English, and emotion datasets demonstrate that I2D enables more reliable automated evaluation for zero-shot TTS.
title Iterate to Differentiate: Enhancing Discriminability and Reliability in Zero-Shot TTS Evaluation
topic Sound
url https://arxiv.org/abs/2603.24430