Flash-CQT paper cuts CQT round-trip time 2-8x on GPU with 30% less memory
drscotthawley · x · 2026-10-01
An arXiv paper (Franchino, Moliner, Välimäki; submitted to ICASSP 2027) introduces Flash-CQT, an exact factorisation of the invertible constant-Q transform for fast GPU computation.
- Problem: CQT represents audio on a log frequency axis; its nonstationary Gabor formulation is exactly invertible, but unequal band coefficient counts make GPU computation inefficient.
- Method: An exact factorisation combines spectral selection, conjugation, windowing, and reordering into a fixed map between one packed Fourier transform and shorter per-band inverse transforms — enabling waveform reconstruction, real adjoints for backpropagation, and bounded-memory streaming.
- Results: On two GPU models, Flash-CQT reduces analysis-synthesis round-trip time by 2-8x versus a baseline computing the same CQT, uses over 30% less peak temporary workspace, and achieves negligible reconstruction error (130 dB SNR) in single precision.
- Takeaway: A practical, compute-efficient front end for spectral analysis and modern audio ML systems.
More from Research
- OSWorld-Science Debuts: 146 Tasks Test How Well VLM Agents Handle Scientific Software — SciAILab · 2026-10-01
- Survey of Attention Evolution: Contextual Memory Becomes the Core of LLM Architecture Design — Zhentao Tan · 2026-10-01
- Hidden Dates in System Prompts Swing LLM Eval Scores by Up to 14% — Mario Sanz-Guerrero · 2026-10-01
- CheatBench Launches to Measure Reward Gaming and Cheating in AI Agents — cais · 2026-10-01
- KLS partially cracked: arXiv paper's core proof ideas generated by AI — burny_tech · 2026-10-01
- Newton's method: when it converges, barely converges, and fails entirely — burny_tech · 2026-10-01