Abstract: A method and apparatus for training a neural network language model, and a method and apparatus for recognizing speech data based on a trained language model are provided. The method of training a language model involves converting, using a processor, training data into error-containing training data, and training a neural network language model using the error-containing training data.

Abstract: An apparatus and method for evaluating quality of an automatic translation is disclosed. An apparatus for evaluating quality of automatic translation includes a converter which converts an automatic translation and a reference translation of an original text to a first distributed representation and a second distributed representation, respectively, using a distributed representation model and a quality evaluator which evaluates quality of automatic translation data based on similarity between the first distributed representation and the second distributed representation.

Abstract: Described are an apparatus and method for generating to generate an acoustic model. The apparatus and method include a processor a processor configured to calculate a noise representation that represents noise data by using a noise model, and generate the acoustic model through training using training noisy speech data, which comprises speech data and the noise data, a string of phonemes corresponding to the speech data, and the noise representation.

Abstract: A speech recognition method and apparatus are provided, in which the speech recognition apparatus may recognize a user feature and a speech recognition environment, determine a speech recognition speed for performing speech recognition based on one of the recognized user feature and the speech recognition environment, and perform the speech recognition based on the determined speech recognition speed.

Abstract: A method and apparatus for training a language model, include generating a first training feature vector sequence and a second training feature vector sequence from training data. The method is configured to perform forward estimation of a neural network based on the first training feature vector sequence, and perform backward estimation of the neural network based on the second training feature vector sequence. The method is further configured to train a language model based on a result of the forward estimation and a result of the backward estimation.

Abstract: A method and apparatus for training a neural network language model, and a method and apparatus for recognizing speech data based on a trained language model are provided. The method of training a language model involves converting, using a processor, training data into error-containing training data, and training a neural network language model using the error-containing training data.

Abstract: A speech recognition method includes receiving a sentence generated through speech recognition, calculating a degree of suitability for each word in the sentence based on a relationship of each word with other words in the sentence, detecting a target word to be corrected among the words in the sentence based on the degree of suitability for each word, and replacing the target word with any one of candidate words corresponding to the target word.

Abstract: A video display method of a user terminal, includes determining whether ambient noise measured when a video is played is in an allowable range, and generating subtitles based on a voice signal included in the video in response to the ambient noise being determined to be out of the allowable range. The method further includes displaying the generated subtitles with the video.

Abstract: A storage controller includes a co-access pattern mining unit configured to detect co-access patterns of data co-accessed during a particular time duration and to generate co-access groups including a plurality of pieces of data complying with the co-access patterns. The storage controller further include a co-access group matching unit configured to select a co-access group matched with read-requested data, among the generated co-access groups, and a data placement unit configured to store the data included in the selected co-access group in a pre-fetch buffer.

Abstract: A speech recognition server and method are provided. The speech recognition server includes: a network connector configured to connect to terminals through a communication network, and a processor configured to calculate times of arrival of a speech sound at each of the terminals using the speech signals received from each of the terminals, calculate distances between a user and the terminals based on the times of arrival of the speech sound to each of the terminals, and recognize a command to control a target by integrating speech recognition results obtained from the speech signals using a weighted value determined based on the distances.

Abstract: An apparatus and a method to estimate a degree of user fatigue to video content are disclosed including a feature value calculating unit and a fatigue degree calculating unit. The feature value calculating unit is configured to calculate feature values corresponding to image and sound features in a video content. The fatigue degree calculating unit is configured to calculate a degree of user fatigue to the video content by applying the feature values to a fatigue degree estimation model.

Abstract: Existing monitors for apnea miss many serious events because they do not adequately distinguish the heart signal in chest impedance from the respiratory signal. Described herein is a respiratory monitoring system and method for improved detection and response to apnea, particularly in a NICU setting but also useful in a home setting. This method filters from the chest impedance the part of the impedance that is caused by the beating of the heart in a human subject, and then identifies in real time significant silence in the filtered chest impedance signal, including determining the probability of apnea. If the probability of apnea exceeds a threshold value, the apneic subject can be stimulated using automated interactions such as a vibrating mattress or air blower.

Abstract: Existing monitors for apnea miss many serious events because they do not adequately distinguish the heart signal in chest impedance from the respiratory signal. Described herein is a respiratory monitoring system and method for improved detection and response to apnea, particularly in a NICU setting but also useful in a home setting. This method filters from the chest impedance the part of the impedance that is caused by the beating of the heart in a human subject, and then identifies in real time significant silence in the filtered chest impedance signal, including determining the probability of apnea. If the probability of apnea exceeds a threshold value, the apneic subject can be stimulated using automated interactions such as a vibrating mattress or air blower.