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

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Mahesh Godavarti
• Venue: arXiv: cs.SD
• Published: 2026-07-22
• Collected: 2026-07-23
• URL: http://arxiv.org/abs/2607.20086v1
• PDF: https://arxiv.org/pdf/2607.20086v1
• DOI: UNKNOWN

Plain-Language Brief

The paper introduces Cumsum-Composable Phase Transport, a new streaming-native temporal layer for keyword spotting in speech tasks. This method uses unitary rotations to project and transport acoustic frames, accumulates them using prefix differences, and applies a gated residual update. It is competitive with compact baselines and provides faster training and lower single-example latency compared to scan-style training kernels.

What the Document Says

• The paper proposes Cumsum-Composable Phase Transport for keyword spotting in streaming speech.
• Each layer projects acoustic frames to complex channels, transports them using unitary rotations, accumulates a finite window using prefix differences, and applies a gated residual update.
• Unitary transport is the key constraint: inverse rotations have norm one.
• The method provides exact batched training with ordinary cumulative sums and exact online inference with one prefix update per frame.
• On Google Speech Commands v2 with 12 labels, mel+cumsum models retain competitive accuracy with compact baselines.

Key Concepts Mentioned

• Cumsum-Composable Phase Transport
• Keyword spotting
• Streaming speech
• Unitary rotations
• Acoustic frames
• Google Speech Commands v2

Claims Made by the Document

• Cumsum-Composable Phase Transport provides competitive accuracy with compact baselines.
• The method trains 1.07x faster than scan-style training kernels.

Evidence Used

• On Google Speech Commands v2, mel+cumsum models reach 97.3% test accuracy versus a 25.6K MelCNNMaxPool baseline at 97.1%.

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.