aiqbench analysis
This study investigates how disordered articulation in dysarthric speech alters internal representations within transformer ASR encoders, revealing that recognition difficulty is encoded primarily in deeper layers. The research demonstrates that parameter-efficient fine-tuning strategies targeting specific mid-to-upper layers can nearly match full encoder adaptation while addressing persistent errors like Mandarin lexical tones. These findings offer a concrete pathway for optimizing low-resource models to handle challenging speech disorders without unnecessary computational overhead.
Analyzing Speech Condition Effects in Dysarthric ASR: A Layer-wise Probing Study
arXiv:2608.01865v1 Announce Type: new Abstract: Automatic speech recognition (ASR) performance degrades sharply on dysarthric speech, yet how disordered articulation reshapes a model's internal representations is underexplored. We present a layer-wise probing analysis of a transformer ASR encoder on Mandarin dysarthric speech under three transcript-matched conditions: original dysarthric speech, speaker conditioned zero-shot TTS resynthesis, and unconditioned TTS. The probes reveal a task-dependent hierarchy: phoneme boundary information stays weak for dysarthric speech at every layer, phoneme identity becomes recoverable toward the upper layers, and recognition difficulty is encoded in the deepest layers. Tone-sensitive evaluation shows Mandarin lexical tone is a persistent error source. Cross-condition similarity divergence grows with depth, indicating that disordered speech affects high-level representations more than low-level acoustic features. Guided by these findings, single-layer LoRA at layer 7 and adaptation on subset layers 5-8 achieve performance within 3.5% and 2.48% relative margins of full encoder adaptation, respectively, while upper-layer adaptation is less effective for dysarthric speech. These findings link representation analysis to parameter-efficient fine-tuning and motivate layer-aware adaptation for low-resource Mandarin dysarthric ASR.
Abstract:Automatic speech recognition (ASR) performance degrades sharply on dysarthric speech, yet how disordered articulation reshapes a model's internal representations is underexplored. We present a layer-wise probing analysis of a transformer ASR encoder on Mandarin dysarthric speech under three transcript-matched conditions: original dysarthric speech, speaker conditioned zero-shot TTS resynthesis, and unconditioned TTS. The probes reveal a task-dependent hierarchy: phoneme boundary information stays weak for dysarthric speech at every layer, phoneme identity becomes recoverable toward the upper layers, and recognition difficulty is encoded in the deepest layers. Tone-sensitive evaluation shows Mandarin lexical tone is a persistent error source. Cross-condition similarity divergence grows with depth, indicating that disordered speech affects high-level representations more than low-level acoustic features. Guided by these findings, single-layer LoRA at layer 7 and adaptation on subset layers 5-8 achieve performance within 3.5% and 2.48% relative margins of full encoder adaptation, respectively, while upper-layer adaptation is less effective for dysarthric speech. These findings link representation analysis to parameter-efficient fine-tuning and motivate layer-aware adaptation for low-resource Mandarin dysarthric ASR.
Computation and Language (cs.CL)
arXiv:2608.01865 [cs.CL]
(or arXiv:2608.01865v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2608.01865 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 263 words
Related coverage
The Most Forbidden Technique is not always forbidden
LessWrong · Jul 18, 2026
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · Oct 8, 2026
Assessing Information Ecosystems : How Governments Can Get Ahead of Hybrid Threats
carnegieendowment.org · Aug 5, 2026
PIRAMID: Progress and Plans
LessWrong · Jul 27, 2026