Google Inc.

Other

Yanzhang He

Tara N Sainath

Ruoming Pang

Trevor D Strohman

Arun Narayanan

Rohit Prakash Prabhavalkar

Ian C McGraw

Qiao Liang

Bo Li

Quan Wang

Chung-Cheng Chiu

Wei Li

Ian Mcgraw

Rajeev Rikhye

Wei Ping Li

Yonghui Wu

Yu Zhang

Kanury Kanishka Rao

Ehsan Variani

David Rybach

Shuo-Yiin Chang

Rami Botros

Cyril Allauzen

Liangliang Cao

Tom O'Malley

Shaojin Ding

David Qiu

Ron J Weiss

Khe Chai Sim

Antoine Jean Bruguier

Gabor Simko

Nathan David Howard

Parisa Haghani

Wei Han

Li Wei

Ke Hu

Qiujia Li

Anton Bakhtin

Jiahui Yu

Lukasz Lew

Weiran Wang

Anmol Gulati

Mirkó Visontai

Qiumin Xu

Shuo-yiin Chang

Alex Seungryong Park

Sepand Mavandadi

Deepti Bhatia

Shivani Agrawal

Phoenix Meadowlark

Oleg Rybakov

Vince Chen

Heguang Liu

Sameer Bidichandani

James Walker

Tsendsuren Munkhdalai

Tara C Sainath

Ian C Mcgraw

A method including receiving a speech recognition result corresponding to a transcription of an utterance spoken by a user. For each sub-word unit in a sequence of hypothesized sub-word units of the speech recognition result, using a confidence estimation module to: obtain a respective confidence embedding associated with the corresponding output step when the corresponding sub-word unit was output from the first speech recognizer; generate a confidence feature vector; generate an acoustic context vector; and generate a respective confidence output score for the corresponding sub-word unit based on the confidence feature vector and the acoustic feature vector received as input by the output layer of the confidence estimation module. The method also includes determining, based on the respective confidence output score generated for each sub-word unit in the sequence of hypothesized sub-word units, an utterance-level confidence score for the transcription of the utterance.

Multi-task learning for end-to-end automated speech recognition confidence and deletion estimation

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

An automatic speech recognition (ASR) system that includes an ASR model, a neural associative memory (NAM) biasing model, and a confidence estimation model (CEM). The ASR model includes an audio encoder configured to encode a sequence of audio frames characterizing a spoken utterance into a sequence of higher-order feature representations, and a decoder configured to receive the sequence of higher-order feature representations and output a final speech recognition result. The NAM biasing model is configured to receive biasing contextual information and modify the sequence of higher-order feature representations based on the biasing contextual information to generate, as output, biasing context vectors. The CEM is configured to compute a confidence of the final speech recognition result output by the decoder. The CEM is connected to the biasing context vectors generated by the NAM biasing model.

Context-aware neural confidence estimation for rare word speech recognition

A method for training a model includes obtaining a plurality of training samples. Each respective training sample of the plurality of training samples includes a respective speech utterance and a respective textual utterance representing a transcription of the respective speech utterance. The method includes training, using quantization aware training with native integer operations, an automatic speech recognition (ASR) model on the plurality of training samples. The method also includes quantizing the trained ASR model to an integer target fixed-bit width. The quantized trained ASR model includes a plurality of weights. Each weight of the plurality of weights includes an integer with the target fixed-bit width. The method includes providing the quantized trained ASR model to a user device.

4-bit conformer with accurate quantization training for speech recognition

A method of text-only and semi-supervised training for deliberation includes receiving training data including unspoken textual utterances that are each not paired with any corresponding spoken utterance of non-synthetic speech, and training a deliberation model that includes a text encoder and a deliberation decoder on the unspoken textual utterances. The method also includes receiving, at the trained deliberation model, first-pass hypotheses and non-causal acoustic embeddings. The first-pass hypotheses is generated by a recurrent neural network-transducer (RNN-T) decoder for the non-causal acoustic embeddings encoded by a non-causal encoder. The method also includes encoding, using the text encoder, the first-pass hypotheses generated by the RNN-T decoder, and generating, using the deliberation decoder attending to both the first-pass hypotheses and the non-causal acoustic embeddings, second-pass hypotheses.

Deliberation by text-only and semi-supervised training

A computer-implemented method includes receiving a sequence of acoustic frames corresponding to an utterance and generating a reference speaker embedding for the utterance. The method also includes receiving a target speaker embedding for a target speaker and generating feature-wise linear modulation (FiLM) parameters including a scaling vector and a shifting vector based on the target speaker embedding. The method also includes generating an affine transformation output that scales and shifts the reference speaker embedding based on the FiLM parameters. The method also includes generating a classification output indicating whether the utterance was spoken by the target speaker based on the affine transformation output.

Optimizing personal VAD for on-device speech recognition

A method includes receiving a sequence of acoustic frames characterizing one or more utterances as input to a multilingual automated speech recognition (ASR) model. The method also includes generating a higher order feature representation for a corresponding acoustic frame. The method also includes generating a hidden representation based on a sequence of non-blank symbols output by a final softmax layer. The method also includes generating a probability distribution over possible speech recognition hypotheses based on the hidden representation generated by the prediction network at each of the plurality of output steps and the higher order feature representation generated by the encoder at each of the plurality of output steps. The method also includes predicting an end of utterance (EOU) token at an end of each utterance. The method also includes classifying each acoustic frame as either speech, initial silence, intermediate silence, or final silence.

Language agnostic multilingual end-to-end streaming on-device ASR system

A method for automatic speech recognition using joint acoustic echo cancellation, speech enhancement, and voice separation includes receiving, at a contextual frontend processing model, input speech features corresponding to a target utterance. The method also includes receiving, at the contextual frontend processing model, at least one of a reference audio signal, a contextual noise signal including noise prior to the target utterance, or a speaker embedding including voice characteristics of a target speaker that spoke the target utterance. The method further includes processing, using the contextual frontend processing model, the input speech features and the at least one of the reference audio signal, the contextual noise signal, or the speaker embedding vector to generate enhanced speech features.

Joint acoustic echo cancelation, speech enhancement, and voice separation for automatic speech recognition

A computer-implemented method of training a streaming speech recognition model that includes receiving, as input to the streaming speech recognition model, a sequence of acoustic frames. The streaming speech recognition model is configured to learn an alignment probability between the sequence of acoustic frames and an output sequence of vocabulary tokens. The vocabulary tokens include a plurality of label tokens and a blank token. At each output step, the method includes determining a first probability of emitting one of the label tokens and determining a second probability of emitting the blank token. The method also includes generating the alignment probability at a sequence level based on the first probability and the second probability. The method also includes applying a tuning parameter to the alignment probability at the sequence level to maximize the first probability of emitting one of the label tokens.

Fast emit low-latency streaming ASR with sequence-level emission regularization utilizing forward and backward probabilities between nodes of an alignment lattice

An ASR model includes a first encoder configured to receive a sequence of acoustic frames and generate a first higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames. The ASR model also includes a second encoder configured to receive the first higher order feature representation generated by the first encoder at each of the plurality of output steps and generate a second higher order feature representation for a corresponding first higher order feature frame. The ASR model also includes a decoder configured to receive the second higher order feature representation generated by the second encoder at each of the plurality of output steps and generate a first probability distribution over possible speech recognition hypothesis. The ASR model also includes a language model configured to receive the first probability distribution over possible speech hypothesis and generate a rescored probability distribution.

Efficient streaming non-recurrent on-device end-to-end model

Techniques disclosed herein are directed towards streaming keyphrase detection which can be customized to detect one or more particular keyphrases, without requiring retraining of any model(s) for those particular keyphrase(s). Many implementations include processing audio data using a speaker separation model to generate separated audio data which isolates an utterance spoken by a human speaker from one or more additional sounds not spoken by the human speaker, and processing the separated audio data using a text independent speaker identification model to determine whether a verified and/or registered user spoke a spoken utterance captured in the audio data. Various implementations include processing the audio data and/or the separated audio data using an automatic speech recognition model to generate a text representation of the utterance. Additionally or alternatively, the text representation of the utterance can be processed to determine whether at least a portion of the text representation of the utterance captures a particular keyphrase. When the system determines the registered and/or verified user spoke the utterance and the system determines the text representation of the utterance captures the particular keyphrase, the system can cause a computing device to perform one or more actions corresponding to the particular keyphrase.

Voice shortcut detection with speaker verification

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for detecting utterances of a key phrase in an audio signal. One of the methods includes receiving, by a key phrase spotting system, an audio signal encoding one or more utterances; while continuing to receive the audio signal, generating, by the key phrase spotting system, an attention output using an attention mechanism that is configured to compute the attention output based on a series of encodings generated by an encoder comprising one or more neural network layers; generating, by the key phrase spotting system and using attention output, output that indicates whether the audio signal likely encodes the key phrase; and providing, by the key phrase spotting system, the output that indicates whether the audio signal likely encodes the key phrase.

Key phrase spotting

A method includes receiving a speech recognition result, and using a confidence estimation module (CEM), for each sub-word unit in a sequence of hypothesized sub-word units for the speech recognition result: obtaining a respective confidence embedding that represents a set of confidence features; generating, using a first attention mechanism, a confidence feature vector; generating, using a second attention mechanism, an acoustic context vector; and generating, as output from an output layer of the CEM, a respective confidence output score for each corresponding sub-word unit based on the confidence feature vector and the acoustic feature vector received as input by the output layer of the CEM. For each of the one or more words formed by the sequence of hypothesized sub-word units, the method also includes determining a respective word-level confidence score for the word. The method also includes determining an utterance-level confidence score by aggregating the word-level confidence scores.

Learning word-level confidence for subword end-to-end automatic speech recognition

Growing community of inventors

Los Altos, CA, United States of America

Yanzhang He