Google Inc.

Salesforce Com, Inc.

Salesforce, Inc.

Weiran Wang

Tara N Sainath

Trevor D Strohman

Rohit Prakash Prabhavalkar

Caiming Xiong

Yingbo Zhou

Rami Botros

Yangzhang He

Sepand Mavandadi

Bhuvana Ramabhadran

Ruoming Pang

Yanzhang He

Ehsan Variani

Neeraj Gaur

David Rybach

Jiahui Yu

Qiao Liang

Ke Hu

Tongzhou Chen

Krzysztof Marcin Choromanski

Anmol Gulati

Ian Mcgraw

Ronny Huang

Charles Caleb Peyser

Shaojin Ding

Ding Zhao

Junwen Bai

Xin Wang

Dongseong Hwang

Robert David

Rina Panigrahy

An automated speech recognition (ASR) model includes a first encoder, a first encoder, a second encoder, and a second decoder. The first encoder receives, as input, a sequence of acoustic frames, and generates, at each of a plurality of output steps, a first higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames. The first decoder receives, as input, the first higher order feature representation generated by the first encoder, and generates a first probability distribution over possible speech recognition hypotheses. The second encoder receives, as input, the first higher order feature representation generated by the first encoder, and generates a second higher order feature representation for a corresponding first higher order feature frame. The second decoder receives, as input, the second higher order feature representation generated by the second encoder, and generates a second probability distribution over possible speech recognition hypotheses.

Unified cascaded encoder ASR model for dynamic model sizes

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 method includes generating, using an audio encoder, a higher-order feature representation for each acoustic frame in a sequence of acoustic frames; generating, using a decoder, based on the higher-order feature representation, a plurality of speech recognition hypotheses, each hypotheses corresponding to a candidate transcription of an utterance and having an associated first likelihood score; generating, using an external language model, for each speech recognition hypothesis, a second likelihood score; determining, using a learnable fusion module, for each speech recognition hypothesis, a set of fusion weights based on the higher-order feature representation and the speech recognition hypothesis; and generating, using the learnable fusion module, for each speech recognition hypothesis, a third likelihood score based on the first likelihood score, the second likelihood score, and the set of fusion weights, the audio encoder and decoder trained using minimum additive error rate training in the presence of the external language model.

Rare word recognition with LM-aware MWER training

Embodiments described herein combine both masked reconstruction and predictive coding. Specifically, unlike contrastive learning, the mutual information between past states and future states are directly estimated. The context information can also be directly captured via shifted masked reconstruction—unlike standard masked reconstruction, the target reconstructed observations are shifted slightly towards the future to incorporate more predictability. The estimated mutual information and shifted masked reconstruction loss can then be combined as the loss function to update the neural model.

Systems and methods for mutual information based self-supervised learning

A computer-implemented method includes receiving a sequence of acoustic frames as input to an automatic speech recognition (ASR) model. Here, the ASR model includes a causal encoder and a decoder. The method also includes generating, by the causal encoder, a first higher order feature representation for a corresponding acoustic frame in the sequence of acoustic frames. The method also includes generating, by the decoder, a first probability distribution over possible speech recognition hypotheses. Here, the causal encoder includes a stack of causal encoder layers each including a Recurrent Neural Network (RNN) Attention-Performer module that applies linear attention.

Optimizing inference performance for conformer

System and methods for identifying a text word from a spoken utterance are provided. An ensemble BPE system that includes a phone BPE system and a character BPE system receives a spoken utterance. Both BPE systems include a multi-level language model (LM) and an acoustic model. The phone BPE system identifies first words from the spoken utterance and determine a first score for each first word. The first words are converted into character sequences. The character BPE model converts the character sequences into second words and determines a second score for each second word. For each word from the first words that matches a word in the second words the first and second scores are combined. The text word is the word with a highest score.

Phone-based sub-word units for end-to-end speech recognition

Growing community of inventors

Palo Alto, CA, United States of America

Weiran Wang