High-Value Synthetic Supervision for Parameter-Efficient Adaptation of a Compact Japanese Speech Model

Excerpt

arXiv:2610.00026v1 Announce Type: cross Abstract: Private domain speech is difficult to collect and redistribute, while compact models need task-specific supervision. We study an auditable synthetic pipeline that maps Japanese care handoffs directly to six-field structured notes. Using 182 synthetic training and development clips, we adapt a 1.47B audio model by full fine-tuning and rank-16 LoRA. On a 39-clip scenario-seed-disjoint synthetic test, an unadapted model obtains a model-judged factua