Handcrafted Topological Feature Boosts Speech Network Noise Robustness

Feng et al. show that adding a single handcrafted persistent-homology feature to deep speech networks lifts accuracy on noisy TIMIT from 84.9% to 88.4%, challenging pure end-to-end learning assumptions.

2026-09-21 ~ 2026-09-21 · 2 related posts

• One topological feature boosts noisy speech recognition accuracy from 84.9% to 88.4% on TIMIT — bravo_abad · 2026-09-21
• One hand-crafted topological feature makes speech neural nets far more noise-robust — bravo_abad · 2026-09-21