Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note3 min read

Cumsum-Composable Phase Transport for Low-Cost Streaming Keyword Spotting

Source

arXiv

arxiv.org/abs/2607.20086v1

Read the source

Source

Plain-Language Brief

A research paper introduces Cumsum-Composable Phase Transport, a new temporal layer designed for keyword spotting in streaming speech. This method projects acoustic frames to complex channels, transports them using learned unitary rotations, accumulates prefix differences, and applies a gated residual update. It offers exact batched training and online inference with one prefix update per frame. The method is found to be competitive in accuracy with compact baselines and faster than scan-style training kernels.

What the Document Says

  • Cumsum-Composable Phase Transport is a streaming-native temporal layer for keyword spotting.
  • It projects acoustic frames to complex channels, transports them using learned unitary rotations, accumulates prefix differences, and applies a gated residual update.
  • The same prefix representation gives exact batched training with ordinary cumulative sums and exact online inference with one prefix update per frame.
  • Unitary transport is the key constraint: inverse rotations have norm one, keeping prefix terms well conditioned while memory is supplied by windows or block readouts.
  • On Google Speech Commands v2 with 12 labels, mel+cumsum models retain competitive accuracy with compact baselines.
  • The strongest single-seed run reaches 97.3% test accuracy.
  • A 51.6K-parameter tied model also reaches 97.3% accuracy.
  • A 24.8K tied model reaches 96.8% accuracy versus 97.1% for a 25.6K MelCNNMaxPool baseline.
  • In a matched cumsum-versus-scan benchmark, cumsum+window gives comparable accuracy and trains 1.07x faster.
  • It reduces single-example latency from 7.09 ms to 5.01 ms on a Tesla T4.

Key Concepts Mentioned

  • Cumsum-Composable Phase Transport
  • Keyword Spotting
  • Streaming Speech
  • Acoustic Frames
  • Complex Channels
  • Unitary Rotations
  • Prefix Differences
  • Gated Residual Update
  • Google Speech Commands v2
  • Mel+cumsum models
  • Compact Baselines
  • MelCNNMaxPool baseline
  • Cumsum+window
  • Scan-style training kernels

Claims Made by the Document

  • Cumsum-Composable Phase Transport is competitive in accuracy with compact baselines.
  • It offers faster training and lower latency compared to scan-style training kernels.

Evidence Used

  • Results from Google Speech Commands v2 experiments and a matched cumsum-versus-scan benchmark.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

State-space sequence models are attractive for streaming speech because they maintain compact recurrent state, but scan-style training kernels can have unfavorable constants for short audio tasks. We study cumsum-composable phase transport, a streaming-native temporal layer for keyword spotting. Each layer projects acoustic frames to complex channels, transports them by learned unitary rotations, accumulates a finite window using prefix differences, and applies a gated residual update. The same prefix representation gives exact batched training with ordinary cumulative sums and exact online inference with one prefix update per frame. Unitary transport is the key constraint: inverse rotations have norm one, keeping prefix terms well conditioned while memory is supplied by windows or block readouts. On Google Speech Commands v2 with 12 labels, mel+cumsum models retain competitive accuracy with compact baselines. The strongest single-seed run reaches 97.3% test accuracy; a 51.6K-parameter tied model also reaches 97.3%, and a 24.8K tied model reaches 96.8% versus 97.1% for a 25.6K MelCNNMaxPool baseline. In a matched cumsum-versus-scan benchmark, cumsum+window gives comparable accuracy, 94.82% versus 94.33%, while training 1.07x faster and reducing single-example latency from 7.09 ms to 5.01 ms on a Tesla T4. These results support cumsum phase transport as a simple low-cost temporal primitive for streaming keyword spotting.