Device for self-adaptive context window partitioning for decentralized speech processing units

Abstract

Device for dynamically dividing a context window of a speech processing unit into hardware-dependent segments, comprising: a) a hardware detection unit (10) for automatically detecting the computing units available in a local computing system and their respective processing capacity, including memory size and computing power; b) a division logic (20) for segmenting the context window into a plurality of sub-segments, wherein the size and content of each sub-segment are adapted to the capabilities of the assigned computing unit; c) an allocation unit (30) for distributing the sub-segments to the respective computing units for parallel processing; and d) a reconstruction unit (40) for coherently combining the parallel processed segment results into a unified processing result.