Prototype-Based Disentanglement for Controllable Dysarthric Speech Synthesis

TL;DR

This paper introduces ProtoDisent-TTS, a prototype-based TTS framework that disentangles speaker identity from dysarthric speech patterns, enabling controllable synthesis and improved robustness for dysarthric speech applications.

Contribution

It presents a novel prototype-based disentanglement method that separates speaker and pathological features in speech synthesis, enhancing controllability and robustness.

Findings

Enables bidirectional transformation between healthy and dysarthric speech.

Improves ASR performance on dysarthric speech.

Provides interpretable and controllable speech representations.

Abstract

Dysarthric speech exhibits high variability and limited labeled data, posing major challenges for both automatic speech recognition (ASR) and assistive speech technologies. Existing approaches rely on synthetic data augmentation or speech reconstruction, yet often entangle speaker identity with pathological articulation, limiting controllability and robustness. In this paper, we propose ProtoDisent-TTS, a prototype-based disentanglement TTS framework built on a pre-trained text-to-speech backbone that factorizes speaker timbre and dysarthric articulation within a unified latent space. A pathology prototype codebook provides interpretable and controllable representations of healthy and dysarthric speech patterns, while a dual-classifier objective with a gradient reversal layer enforces invariance of speaker embeddings to pathological attributes. Experiments on the TORGO dataset demonstrate that this design enables bidirectional transformation between healthy and dysarthric speech, leading to consistent ASR performance gains and robust, speaker-aware speech reconstruction.