Patent · US Active

Universal acoustic modeling using neural mixture models

US11676006B2 · kind B2 · utility

1Cited by
0References
21Claims
0Family size

Assignee

Inventors

Key dates

Filing dateMay 16, 2019
Grant dateJun 13, 2023
Priority date
Expiry dateDec 2, 2041

Classification

  • Technology area (CPC G)Physics
  • CPC primaryG10L15/32
  • WIPO fieldComputer technology
  • WIPO sectorElectrical engineering

Abstract

According to some embodiments, a universal modeling system may include a plurality of domain expert models to each receive raw input data (e.g., a stream of audio frames containing speech utterances) and provide a domain expert output based on the raw input data. A neural mixture component may then generate a weight corresponding to each domain expert model based on information created by the plurality of domain expert models (e.g., hidden features and/or row convolution). The weights might be associated with, for example, constrained scalar numbers, unconstrained scaler numbers, vectors, matrices, etc. An output layer may provide a universal modeling system output (e.g., an automatic speech recognition result) based on each domain expert output after being multiplied by the corresponding weight for that domain expert model.

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.