Abstract: Disclosed is a method for detecting a speech segment, which is performed by a computing device. The method may include: detecting a start point of a speech segment in an audio signal; and detecting an end point of the speech segment based on an offset threshold which is dynamically changed, and the dynamically changed offset threshold may be based on a length of the speech segment.

Abstract: Disclosed is a method for determining a content associated with a voice signal, which is performed by a computing device. The method may include converting a voice signal and generating text information. The method may include determining a plurality of target word candidates. The method may include determining a target word among the plurality of target word candidates based on a comparison between the plurality of target word candidates and the generated text information. The method may also include determining a content associated with the target word.

Abstract: Disclosed is a voice recognition method based on a neural network, performed by one or more processors of a computing device. The voice recognition method includes generating first text data based on a first language based on voice information. The method includes detecting problem text among the generated first text data. The method includes detecting a second language corresponding to the detected problem text. The method includes generating second text data based on the detected second language, in which the second language is different from the first language.

Abstract: Disclosed is a method for detecting a voice from audio data, performed by a computing device according to an exemplary embodiment of the present disclosure. The method includes obtaining audio data; generating image data based on a spectrum of the obtained audio data; analyzing the generated image data by utilizing a pre-trained neural network model; and determining whether an automated response system (ARS) voice is included in the audio data, based on the analysis of the image data.

Abstract: Disclosed is a method for error detection, performed by one or more processors of a computing device according to an example embodiment of the present disclosure. The method includes evaluating an error rate for a sentence to be evaluated, in a first language unit. the method includes evaluating an error rate in a second language unit which is smaller than the first language unit, based on the first language unit error.

Abstract: Disclosed is a method for speech recognition performed by one or more processors of a computing device. The method includes inputting voice information into an encoder to extract a first feature vector and calculating a first loss function. The method includes inputting the first feature vector extracted from the encoder to a first decoder to perform prediction on the voice information, calculating a second loss function, and extracting a second feature vector. The method includes inputting a second feature vector extracted from the first decoder to a second decoder to perform grapheme-based prediction, and calculating a third loss function. The method includes training at least one of the encoder, the first decoder, or the second decoder based on the first loss function, the second loss function, and the third loss function.

Abstract: Disclosed is a method for generating data, the method is performed by one or more processors of a computing device. The method may include: segmenting text data generated based on speech information into a token unit; generating a first feature vector based on the text data segmented into the token unit, and generating a first label vector corresponding to the generated first feature vector, and generating a second feature vector and a second label vector by performing mix-up for each of the generated first feature vector and the generated first label vector.

Abstract: Disclosed is a speech recognition method performed by one or more processors of a computing device, the speech recognition method including: performing first speech recognition on voice information to obtain first text information on the voice information; receiving feedback regarding the first text information; and generating final text information for the voice information based on the received feedback, in which the first speech recognition includes real-time speech recognition, and is performed through a neural network model of a first structure.

Abstract: Disclosed is a method for dialogued summarization with word graphs, which is performed by one or more processors of a computing device. The method may include: generating a word graph based on information on a dialogue which is a summary target; extracting at least one keyword based on the generated word graph; generating a plurality of candidate summary sentences based on the generated word graph; and calculating a score associated with at least one keyword for each of the plurality of candidate summary sentences, and selecting at least one of the plurality of candidate summary sentences based on the calculated score.