Emformer: cached/block-parallel memory attention for streaming recognition

What changed

Combined summarized history memory, left-context KV caching and parallel block training for low-latency acoustic transformers.

Scientific scope

Combined summarized history memory, left-context KV caching and parallel block training for low-latency acoustic transformers. Inherits memory transformers, blockwise streaming and TransformerXL. Bounded memory summaries are not lossless all-history attention; latency/word-error gains are original model/data/hardware observations.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits memory transformers, blockwise streaming and TransformerXL. Bounded memory summaries are not lossless all-history attention; latency/word-error gains are original model/data/hardware observations.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Emformer: Efficient Memory Transformer Based Acoustic Model For Low Latency Streaming Speech Recognition

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction