CosyAccent: Duration-Controllable Accent Normalization Using Source-Synthesis Training Data

TL;DR

This paper introduces CosyAccent, a non-autoregressive accent normalization model trained with source-synthesis data, achieving natural and duration-controlled speech output without using real L2 speech data.

Contribution

The paper presents a novel source-synthesis training approach and a non-autoregressive model that improves accent normalization by enhancing naturalness and duration control without real L2 data.

Findings

Outperforms baselines trained on real data in naturalness.

Achieves better content preservation.

Provides explicit duration control.

Abstract

Accent normalization (AN) systems often struggle with unnatural outputs and undesired content distortion, stemming from both suboptimal training data and rigid duration modeling. In this paper, we propose a "source-synthesis" methodology for training data construction. By generating source L2 speech and using authentic native speech as the training target, our approach avoids learning from TTS artifacts and, crucially, requires no real L2 data in training. Alongside this data strategy, we introduce CosyAccent, a non-autoregressive model that resolves the trade-off between prosodic naturalness and duration control. CosyAccent implicitly models rhythm for flexibility yet offers explicit control over total output duration. Experiments show that, despite being trained without any real L2 speech, CosyAccent achieves significantly improved content preservation and superior naturalness compared to strong baselines trained on real-world data.