Tencent America LLC

Jia Cui

Dong Yu

Chao Weng

Chengzhu Yu

Jun Wang

Dan Su

Guangsen Wang

Peidong Wang

Methods and apparatuses are provided for performing sequence to sequence (Seq2Seq) speech recognition training performed by at least one processor. The method includes acquiring a training set comprising a plurality of pairs of input data and target data corresponding to the input data, encoding the input data into a sequence of hidden states, performing a connectionist temporal classification (CTC) model training based on the sequence of hidden states, performing an attention model training based on the sequence of hidden states, and decoding the sequence of hidden states to generate target labels by independently performing the CTC model training and the attention model training.

Multi-task training architecture and strategy for attention-based speech recognition system

A method and apparatus are provided that analyzing sequence-to-sequence data, such as sequence-to-sequence speech data or sequence-to-sequence machine translation data for example, by minimum Bayes risk (MBR) training a sequence-to-sequence model and within introduction of applications of softmax smoothing to an N-best generation of the MBR training of the sequence-to-sequence model.

N-best softmax smoothing for minimum bayes risk training of attention based sequence-to-sequence models

A method of attention-based end-to-end (A-E2E) automatic speech recognition (ASR) training, includes performing cross-entropy training of a model, based on one or more input features of a speech signal, determining a posterior probability vector at a time of a first wrong token among one or more output tokens of the model of which the cross-entropy training is performed, and determining a loss of the first wrong token at the time, based on the determined posterior probability vector. The method further includes determining a total loss of a training set of the model of which the cross-entropy training is performed, based on the determined loss of the first wrong token, and updating the model of which the cross-entropy training is performed, based on the determined total loss of the training set.

Token-wise training for attention based end-to-end speech recognition

Methods and apparatuses are provided for performing acoustic to word (A2W) speech recognition training performed by at least one processor. The method includes initializing, by the at least one processor, one or more first layers of a neural network with phone based Connectionist Temporal Classification (CTC), initializing, by the at least one processor, one or more second layers of the neural network with grapheme based CTC, acquiring, by the at least one processor, training data and performing, by the at least one processor, A2W speech recognition training based the initialized one or more first layers and one or more second layers of the neural network using the training data.

Multistage curriculum training framework for acoustic-to-word speech recognition

A method for selecting an optimal language model weight (LMW) used to perform automatic speech recognition, including decoding test audio into a lattice using a language model; analyzing the lattice using a first LMW of a plurality of LMWs to determine a first plurality of best paths; analyzing the lattice using a second LMW of the plurality of LMWs to determine a second plurality of best paths; determining a first best path change rate (BCPR) corresponding to the first LMW based on a number of best path changes between the first plurality of best paths and the second plurality of best paths; and determining the first LMW to be the optimal LMW based on the first BCPR being a lowest BCPR from among a plurality of BCPRs corresponding to the plurality of LMWs.

Best path change rate for unsupervised language model weight selection

A method of attention-based end-to-end (E2E) automatic speech recognition (ASR) training, includes performing cross-entropy training of a model, based on one or more input features of a speech signal, performing beam searching of the model of which the cross-entropy training is performed, to generate an n-best hypotheses list of output hypotheses, and determining a one-best hypothesis among the generated n-best hypotheses list. The method further includes determining a character-based gradient and a word-based gradient, based on the model of which the cross-entropy training is performed and a loss function in which a distance between a reference sequence and the determined one-best hypothesis is maximized, and performing backpropagation of the determined character-based gradient and the determined word-based gradient to the model, to update the model.

Large margin training for attention-based end-to-end speech recognition

Methods and apparatuses are provided for performing end-to-end speech recognition training performed by at least one processor. The method includes receiving, by the at least one processor, one or more input speech frames, generating, by the at least one processor, a sequence of encoder hidden states by transforming the input speech frames, computing, by the at least one processor, attention weights based on each of the sequence of encoder hidden states and a current decoder hidden state, performing, by the at least one processor, a decoding operation based on a previous embedded label prediction information and a previous attentional hidden state information generated based on the attention weights; and generating a current embedded label prediction information based on a result of the decoding operation and the attention weights.

Growing community of inventors

Bellevue, WA, United States of America

Jia Cui