WRSWorld Research StateContribute
WORLD RESEARCH STATE

Hwang et al.: fully shared depth decoding for memory-efficient speech synthesis

Contribution grade

On-device speech-generation systems · Human-designed research method or resource

Reference score55,000Provisional editorial estimate of this exact contribution; its grade is assessed separately.

What changed

Separated temporal and codec-depth processing and reused a fully shared, stage-conditioned depth decoder across quantization levels, combining it with a fixed-window cache for streaming semantic-to-audio detokenization.

Scientific scope

Separated temporal and codec-depth processing and reused a fully shared, stage-conditioned depth decoder across quantization levels, combining it with a fixed-window cache for streaming semantic-to-audio detokenization. Moshi's temporal/depth factorization and DiT-style conditioning are inherited. Delta removes per-codebook projections, not invention of depth decoding. Constant memory concerns the fixed-window synthesis component, not total assets, output storage or the entire assistant. AMX latency/memory and deployment are source-reported.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Moshi's temporal/depth factorization and DiT-style conditioning are inherited. Delta removes per-codebook projections, not invention of depth decoding. Constant memory concerns the fixed-window synthesis component, not total assets, output storage or the entire assistant. AMX latency/memory and deployment are source-reported.

Contributors

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