Google Inc.

Asml Netherlands B.v.

Antoine Jean Bruguier

Rohit Prakash Prabhavalkar

Tara N Sainath

Wenjin Shao

Yu Long Cao

Golan Pundak

Luoqi Chen

Bryan Christopher Horling

Ouais Alsharif

Ian C McGraw

Ke Hu

Ashutosh Shukla

Jun Ye

Bo Li

Francoise Beaufays

Ruoming Pang

Fuchun Peng

Arun Narayanan

Yanzhang He

Wei Li

Shuo-yiin Chang

Song Lan

Ian C Mcgraw

Two-pass automatic speech recognition (ASR) models can be used to perform streaming on-device ASR to generate a text representation of an utterance captured in audio data. Various implementations include a first-pass portion of the ASR model used to generate streaming candidate recognition(s) of an utterance captured in audio data. For example, the first-pass portion can include a recurrent neural network transformer (RNN-T) decoder. Various implementations include a second-pass portion of the ASR model used to revise the streaming candidate recognition(s) of the utterance and generate a text representation of the utterance. For example, the second-pass portion can include a listen attend spell (LAS) decoder. Various implementations include a shared encoder shared between the RNN-T decoder and the LAS decoder.

Two-pass end to end speech recognition

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for implementing a compressed recurrent neural network (RNN). One of the systems includes a compressed RNN, the compressed RNN comprising a plurality of recurrent layers, wherein each of the recurrent layers has a respective recurrent weight matrix and a respective inter-layer weight matrix, and wherein at least one of recurrent layers is compressed such that a respective recurrent weight matrix of the compressed layer is defined by a first compressed weight matrix and a projection matrix and a respective inter-layer weight matrix of the compressed layer is defined by a second compressed weight matrix and the projection matrix.

Compressed recurrent neural network models

A method includes receiving audio data encoding an utterance spoken by a native speaker of a first language, and receiving a biasing term list including one or more terms in a second language different than the first language. The method also includes processing, using a speech recognition model, acoustic features derived from the audio data to generate speech recognition scores for both wordpieces and corresponding phoneme sequences in the first language. The method also includes rescoring the speech recognition scores for the phoneme sequences based on the one or more terms in the biasing term list, and executing, using the speech recognition scores for the wordpieces and the rescored speech recognition scores for the phoneme sequences, a decoding graph to generate a transcription for the utterance.

Phoneme-based contextualization for cross-lingual speech recognition in end-to-end models

A method of biasing speech recognition includes receiving audio data encoding an utterance and obtaining a set of one or more biasing phrases corresponding to a context of the utterance. Each biasing phrase in the set of one or more biasing phrases includes one or more words. The method also includes processing, using a speech recognition model, acoustic features derived from the audio data and grapheme and phoneme data derived from the set of one or more biasing phrases to generate an output of the speech recognition model. The method also includes determining a transcription for the utterance based on the output of the speech recognition model.

Contextual biasing for speech recognition

Contextual biasing for speech recognition using grapheme and phoneme data

Methods and apparatus are disclosed for resolving multiple interpretations of an ambiguous temporal term of a resource to a subset of the multiple interpretations. In some implementations, a group of one or more messages is identified, an ambiguous temporal term of the messages determined, additional content of the messages determined, and multiple interpretations of the ambiguous temporal term resolved to a subset based on the additional content.

Date and/or time resolution

Methods, systems, and apparatus, including computer programs encoded on computer storage medium, for implementing a pronunciation dictionary that stores entity name pronunciations. In one aspect, a method includes actions of receiving audio data corresponding to an utterance that includes a command and an entity name. Additional actions may include generating, by an automated speech recognizer, an initial transcription for a portion of the audio data that is associated with the entity name, receiving a corrected transcription for the portion of the utterance that is associated with the entity name, obtaining a phonetic pronunciation that is associated with the portion of the audio data that is associated with the entity name, updating a pronunciation dictionary to associate the phonetic pronunciation with the entity name, receiving a subsequent utterance that includes the entity name, and transcribing the subsequent utterance based at least in part on the updated pronunciation dictionary.

Learning personalized entity pronunciations

Embodiments of the present invention describe methods of selecting a subset of test patterns from an initial larger set of test patterns for calibrating a computational lithography model. An example method comprises: generating an information matrix for the initial larger set of test patterns, wherein the terms of the information matrix comprise one or more identified model parameters that represent a lithographic process response; and, executing a selection algorithm using terms of the information matrix to select the subset of test patterns that effectively determines values of the identified model parameters that contribute significantly in the lithographic process response, wherein the subset of test patterns characteristically represents the initial larger set of test patterns. The selection algorithm explores coverage relationships existing in the initial larger set of test patterns.

Information matrix creation and calibration test pattern selection based on computational lithography model parameters

A method for determining an image of a mask pattern in a resist coated on a substrate, the method including determining an aerial image of the mask pattern at substrate level; and convolving the aerial image with at least two orthogonal convolution kernels to determine a resist image that is representative of the mask pattern in the resist.

Harmonic resist model for use in a lithographic apparatus and a device manufacturing method

The preset invention provides methods, systems and computer program product for selection of an optimum set of patterns to calibrate a lithography model so that the model can predict imaging performance of a lithography apparatus/system more accurately and reliably without being prohibitively expensive in terms of using computational and metrology resources and time. The method is based on modeling sensitivity of the calibration patterns to measurement noise. In one aspect of the present invention, a method is disclosed, comprising: identifying a model of at least a portion of a lithographic process; identifying a set of patterns for calibrating the model; and, estimating measurement noise associated with the set of patterns.

Growing community of inventors

Milpitas, CA, United States of America

Antoine Jean Bruguier