EmpDPO: Enabling Emphasis Control in TTS via Direct Preference Optimization

arXiv:2610.39427 · cs.SD, eess.AS · Submitted 2026-10-29 · Read on arXiv

cs.SD, eess.AS

Submitted: 2026-10-29

Updated: 2026-10-29

Comments: 5 pages, 3 figures. Submitted to ICASSP 2027. Audio samples: https://chaonan99.github.io/empdpo-demo/

License: http://creativecommons.org/licenses/by/4.0/

The gist: Word-level emphasis conveys meaning that text alone cannot express and is therefore essential for a practical text-to-speech (TTS) system.

Abstract

Word-level emphasis conveys meaning that text alone cannot express and is therefore essential for a practical text-to-speech (TTS) system. In this work, we study how to add emphasis control post hoc to a pretrained TTS model that has never been trained with emphasis annotations. We present EmpDPO, a simple, efficient, and effective two-stage recipe that adds emphasis control to a multilingual zero-shot diffusion TTS model through supervised fine-tuning followed by direct preference optimization (DPO). The training signals for both stages come from an off-the-shelf multilingual emphasis detector, so that no manual emphasis annotation or task-specific reward design is needed. We find that DPO can substantially strengthen emphasis control with only a few thousand preference pairs and training steps. Objective and subjective evaluations against strong public systems show that our model produces perceptible emphasis while preserving naturalness and overall quality.

Related papers