Method to speed up the execution of neural networks for voice recognition and voice related mechanism.

摘要

Method for accelerating neural network execution (4) in a system of speech recognition for recognizing words contained in a subset of a general vocabulary of words that the system itself is capable of recognizing, said neural network (4) comprising a series of computing units organized in levels, including at least one hidden level (12) and an output level (14), the computing units (Hj) of said hidden level (12) being connected to the computing units ( Ni) of said output level (14) via weighted connections (Wij), corresponding said computing units (Ni) of said output level (14) with acoustic-phonetic units (2) of said general vocabulary, characterized that comprises the following steps: - determining a subset of acoustic-phonetic necessary for recognizing all the words contained in said general vocabulary subset units; - removing the neural network (4) all afferent weighted connections (Wij) to computing units (Ni) of said output level (14) that correspond to acoustic-phonetic units not contained in said subset of acoustic units -fonéticas previously determined, thereby obtaining a compacted neural network (4 ') optimized for recognition of the words contained in said general vocabulary subset; - executing, at each moment in time, only said compacted neural network (4 '').