Training Data Augmentation for Dysarthric Automatic Speech Recognition by Text-to-Dysarthric-Speech Synthesis

TL;DR

This work tackles DASR data scarcity by training a diffusion-based TTDS system (Grad-TTS) from scratch on TORGO dysarthric data to generate dysarthric-like mel-spectrograms, which are converted to audio via HiFi-GAN. The synthesized data are used to finetune large ASR foundation models (Whisper), exploring both data amount and SpecAugment, yielding substantial WER reductions, especially for severely dysarthric speech. The results demonstrate a path to state-of-the-art DASR without relying on matched control speech, highlighting the practical impact of TTDS augmentation for PwD-enabled AAC and home environments. The approach shows that diffusion-based TTDS can capture dysarthric characteristics and meaningfully improve adaptation of large-scale ASR models to DASR tasks, with performance scaling with synthetic data volume and severity considerations.

Abstract

Automatic speech recognition (ASR) research has achieved impressive performance in recent years and has significant potential for enabling access for people with dysarthria (PwD) in augmentative and alternative communication (AAC) and home environment systems. However, progress in dysarthric ASR (DASR) has been limited by high variability in dysarthric speech and limited public availability of dysarthric training data. This paper demonstrates that data augmentation using text-to-dysarthic-speech (TTDS) synthesis for finetuning large ASR models is effective for DASR. Specifically, diffusion-based text-to-speech (TTS) models can produce speech samples similar to dysarthric speech that can be used as additional training data for fine-tuning ASR foundation models, in this case Whisper. Results show improved synthesis metrics and ASR performance for the proposed multi-speaker diffusion-based TTDS data augmentation for ASR fine-tuning compared to current DASR baselines.