One-step flow matching-based speech enhancement via interpolated mean-vector-field self-distillation
One-step flow matching-based speech enhancement via interpolated mean-vector-field self-distillation Seonggyu Lee Gwangju Institute of Science and Technology (GIST), Republic of Korea Technical Report: Seongqjini-TR-2026-001 First posted: September 6, 2026 Current version: v0.2 — September 9, 2026 PDF | BibTeX Abstract Keywords: speech enhancement; flow matching; one-step generation; knowledge distillation; mean vector field 1. Introduction Speech … Read more