DLM-AN: Controllable Accent Normalization via Discrete Diffusion

This repository contains the pretrained checkpoints for DLM-AN, the inference release for the paper Controllable Accent Normalization via Discrete Diffusion (Interspeech 2026).

DLM-AN (Diffusion Language Model for Accent Normalization) is a controllable accent normalization system built on masked discrete diffusion over self-supervised speech tokens. It provides tunable accent strength through a Common Token Predictor that selectively reuses source tokens, and includes a flow-matching Duration Ratio Predictor to better match native rhythm.

• Project page: https://P1ping.github.io/dlman-demo
• Code: https://github.com/P1ping/DLM-AN