Samsung Electronics Co., Ltd.

Min-Joong Lee

Hwidong Na

Tae Gyoon Kang

Hodong Lee

Jihyun Lee

Jinwoo Park

YoungMin Kim

Young-Seok Kim

Hoshik Lee

YoungSang Choi

Seongmin Ok

A decoding method includes receiving an input sequence corresponding to an input speech at a current time; and in a neural network (NN) for speech recognition, generating an encoded vector sequence by encoding the input sequence, determining reuse tokens from candidate beams of two or more previous times by comparing the candidate beams of the previous times, and decoding one or more tokens subsequent to the reuse tokens based on the reuse tokens and the encoded vector sequence.

Method and apparatus with decoding in neural network for speech recognition

A processor-implemented utterance time estimation method includes: determining a plurality of attention weight matrices using an attention-based sequence-to-sequence model; selecting an attention weight matrix from the plurality of attention weight matrices; and estimating an utterance time corresponding to an output sequence based on the selected attention weight matrix.

Method and apparatus with utterance time estimation

A decoding method and apparatus in an artificial neural network for speech recognition. The decoding method in the artificial neural network for speech recognition includes performing a first decoding task of decoding a feature including speech information and at least one token recognized up to current time, using a shared decoding layer included in the artificial neural network, performing a second decoding task of decoding the at least one token, using the shared decoding layer, and determining an output token to be recognized subsequent to the at least one token based on a result of the first decoding task and a result of the second decoding task.

Decoding method and apparatus in artificial neural network for speech recognition

A speech recognition method includes adding a preset special sequence to a front end of an input sequence that corresponds to an input utterance of a speaker, recognizing the preset special sequence and the input sequence, and recognizing the input sequence based on the preset special sequence and a speech recognition result obtained by recognizing the preset special sequence and the input sequence.

Method and apparatus for speech recognition

A method and apparatus for improving the quality of an attention-based sequence-to-sequence model. The method includes determining an output sequence corresponding to an input sequence based on an attention-based sequence-to-sequence model, selecting at least one target attention head from among a plurality of attention heads, detecting at least one error output token among output tokens constituting the output sequence based on the target attention head, and correcting the output sequence based on the error output token.

Method and apparatus for improving quality of attention-based sequence-to-sequence model

A real-time processor-implemented translation method and apparatus is provided. The real-time translation method includes receiving a content, determining a delay time for real-time translation based on a silence interval of the received content and an utterance interval of the received content, generating a translation result by translating a language used in the received content, and synthesizing the translation result and the received content.

Method and apparatus with real-time translation

A method for determining an output token includes predicting a first probability of each of candidate output tokens of a first model, predicting a second probability of each of the candidate output tokens of a second model interworking with the first model, adjusting the second probability of each of the candidate output tokens based on the first probability, and determining the output token among the candidate output tokens based on the first probability and the adjusted second probability.

Method and apparatus for determining output token

A processor-implemented decoding method in a first neural network is provided. The method predicts probabilities of candidates of an output token based on at least one previously input token, determines the output token among the candidates based on the predicted probabilities; and determines a next input token by selecting one of the output token and a pre-defined special token based on a determined probability of the output token.

Method and apparatus with speech recognition

A processor implemented natural language processing method and apparatus are provided. The natural language processing method includes converting a natural language phrase into a token vector, calculating a repetition count of the token vector, and generating an input vector by encoding the token vector based on the calculated repetition count and a position of the token vector.

Natural language processing method and apparatus

A sequence processing method and apparatus are provided. The sequence processing method includes determining a word of a first R-node corresponding to a root node based on an input sequence, generating first I-nodes that are connected to the first R-node and include relative position information with respect to the word of the first R-node, determining a word of a second R-node to correspond to each of the first I-nodes, and determining an output sequence corresponding to the input sequence based on the determined words.

Method and apparatus for processing sequence

A processor-implemented classification method includes: determining a first probability vector including a first probability, for each of a plurality of classes, resulting from a classification of an input with respect to the classes; determining, based on the determined first probability vector, whether one or more of the classes represented in the first probability vector are confusing classes; adjusting, in response to one or more of the classes being the confusing classes, the determined first probability vector based on a first probability of each of the confusing classes and a maximum value of the first probabilities; determining a second probability vector including a second probability, for each of the classes, resulting from another classification of the input with respect to the classes; and performing classification on the input based on a result of a comparison between the determined second probability vector and the adjusted first probability vector.

Apparatus and method with classification

A language processing method and apparatus is disclosed. A language processing apparatus using a neural network may obtain context information from a source text using a neural network-based encoder, generate a prefix token from the context information using a neural network-based main decoder, generate a token sequence including at least two successive tokens sequentially following the prefix token using a skip model in response to the prefix token satisfying a preset condition, and indicate a target text in which the prefix token and the token sequence are combined as an inference result with respect to the source text.

Language processing method and apparatus

A machine translation method and a machine translation apparatus using a neural network model are provided. The machine translation apparatus extracts information associated with a keyword from a source sentence, obtains a supplement sentence associated with the source sentence based on the extracted information associated with the keyword, acquires a first vector value from the source sentence and a second vector value from the supplement sentence using neural network model-based encoders, and outputs a target sentence corresponding to a translation of the source sentence based on any one or any combination of the first vector value and the second vector value using a neural network model-based decoder.

Growing community of inventors

Suwon-si, South Korea

Min-Joong Lee