Electronics and Telecommunications Research Institute

Industry-University Cooperation Foundation Hanyang University

Young Jik Lee

Sang Hun Kim

Mu Yeol Choi

Dong Hyun Kim

Bon Ki Koo

Seung Wook Lee

Ji Young Park

Ji Hyung Lee

Ho Won Kim

Chang Woo Chu

Jin Seo Kim

Seung Yun

Jun Soo Park

Il Kyu Park

Min Kyu Lee

Bon Woo Hwang

Seung Hi Kim

Soo Jong Lee

Ki Hyun Kim

Chung Hee Lee

Myung Gil Jang

Hoon Chung

Ji Hyun Wang

Euee Seon Jang

Seong Jae Lim

Kap Kee Kim

Hyo Jung Oh

Bo Youn Kim

Jeung Chul Park

Jae Woo Yang

Hyeon Jin Kim

Hyung Bae Jeon

Yoon-Seok Choi

Chang Ki Lee

Jeong Se Kim

Byoungjun Kim

Yun Young Kim

Jaebum Jun

Kyoung Soo Son

Seung Shin Oh

Giseok Son

Eung Kyeu Kim

Young Joo Suh

Seong Jae Lin

Daiyong Kim

Provided are an apparatus and method for selecting a speaker by using smart glasses. The apparatus includes a camera configured to capture a front angle video of a user and track guest interpretation interlocutors in the captured video, smart glasses configured to display a virtual space map image including the guest interpretation interlocutors tracked through the camera, a gaze-tracking camera configured to select a target person for interpretation by tracking a gaze of the user so that a guest interpretation interlocutor displayed in the video may be selected, and an interpretation target processor configured to provide an interpretation service in connection with the target person selected through the gaze-tracking camera.

Apparatus and method for selecting speaker by using smart glasses

Provided is a method of providing an automatic speech translation service. The method includes, by an automatic speech translation device of a user, searching for and finding a nearby automatic speech translation device based on strength of a signal for wireless communication, exchanging information for automatic speech translation with the found automatic speech translation device, generating a list of candidate devices for the automatic speech translation using the automatic speech translation information and the signal strength, and connecting to a candidate device having a greatest variation of the signal strength among devices in the generated list.

Method and apparatus for providing automatic speech translation service in face-to-face situation

A speech recognition method capable of automatic generation of phones according to the present invention includes: unsupervisedly learning a feature vector of speech data; generating a phone set by clustering acoustic features selected based on an unsupervised learning result; allocating a sequence of phones to the speech data on the basis of the generated phone set; and generating an acoustic model on the basis of the sequence of phones and the speech data to which the sequence of phones is allocated.

Speech recognition system and method

Provided are an automatic interpretation system and method for generating a synthetic sound having characteristics similar to those of an original speaker's voice. The automatic interpretation system for generating a synthetic sound having characteristics similar to those of an original speaker's voice includes a speech recognition module configured to generate text data by performing speech recognition for an original speech signal of an original speaker and extract at least one piece of characteristic information among pitch information, vocal intensity information, speech speed information, and vocal tract characteristic information of the original speech, an automatic translation module configured to generate a synthesis-target translation by translating the text data, and a speech synthesis module configured to generate a synthetic sound of the synthesis-target translation.

Automatic interpretation system and method for generating synthetic sound having characteristics similar to those of original speaker's voice

The present invention relates to a method of creating an animatable digital clone includes receiving input multi-view images of an actor captured by at least two cameras and reconstructing a three-dimensional appearance therefrom, accepting shape information selectively based on a probability of photo-consistency in the input multi-view images obtained from the reconstruction and transferring a mesh topology of a reference human body model onto a shape of the actor obtained from the reconstruction. The method further includes generating an initial human body model of the actor via transfer of the mesh topology utilizing sectional shape information of the actor's joints, and generating a genuine human body model of the actor from learning genuine behavioral characteristics of the actor by applying the initial human body model to multi-view posture learning images where performance of a predefined motion by the actor is recorded.

Method of creating animatable digital clone from multi-view images

An apparatus for 3D mesh compression based on quantization, includes a data analyzing unit () for decomposing data of an input 3D mesh model into vertices information () property information () representing property of the 3D mesh model, and connectivity information () between vertices constituting the 3D mesh model: and a mesh model quantizing unit () for producing quantized vertices and property information of the 3D mesh model by using the vertices, property and connectivity information (). Further, the apparatus for 3D mesh compression based on quantization includes a decision bit encoding unit () for calculating a decision bit by using the quantized connectivity information and then encoding the quantized vertex information, property information and connectivity information () by using the decision bit.

Method and apparatus for real time 3D mesh compression, based on quanitzation

Provided are an automatic speech translation system and a method for obtaining accurate translation performance with a simple structure. Because input and output sentences are written in different languages, automatic speech translation requires techniques for processing different languages. Repetition of text processing like morpheme analysis or sentence parsing in conventional automatic speech translation can complicate the overall translation process. Meanwhile, although input and output sentences are written in different languages, they have to have the same meaning and a corresponding sentence form and words. Accordingly, the corresponding words and sentence forms of the two languages can be expressed with a simple structure and utilized in the automatic speech translation process, thereby maintaining consistency during the process and avoiding unnecessary process repetition, which reduces errors and improves performance.

Automatic translation method and system based on corresponding sentence pattern

Provided are an apparatus and method for speech segment detection, and a system for speech recognition. The apparatus is equipped with a sound receiver and an image receiver and includes: a lip motion signal detector for detecting a motion region from image frames output from the image receiver, applying lip motion image feature information to the detected motion region, and detecting a lip motion signal; and a speech segment detector for detecting a speech segment using sound frames output from the sound receiver and the lip motion signal detected from the lip motion signal detector. Since lip motion image information is checked in a speech segment detection process, it is possible to prevent dynamic noise from being misrecognized as speech.

Apparatus and method for speech segment detection and system for speech recognition

An apparatus and method for adjusting the friendliness of a synthesized speech and thus generating synthesized speech of various styles in a speech synthesis system are provided. The method includes the steps of defining at least two friendliness levels; storing recorded speech data of sentences, the sentences being made up according to each of the friendliness levels; extracting at least one of prosodic characteristics for each of the friendliness levels from the recorded speech data, said prosodic characteristics including at least one of a sentence-final intonation type, boundary intonation types of intonation phrases in the sentence, and an average value of Fof the sentence, with respect to the recorded speech data; and generating a prosodic model for each of the friendliness levels by statistically modeling the at least one of the prosodic characteristics.

Method of generating a prosodic model for adjusting speech style and apparatus and method of synthesizing conversational speech using the same

An apparatus and method for efficiently constructing learning data required in statistical methodology used in information retrieval, information extraction, translation, natural language processing, etc. are provided. The method includes the steps of: generating learning models by performing machine learning with respect to learning data; attaching tags to a raw corpus automatically by using the generated learning models to thereby generate learning data candidates; calculating confidence scores of the generated learning data candidates, and then selecting a learning data candidate using the confidence scores; and allowing a user to correct an error in the selected learning data candidate through an interface and adding the error-corrected learning data candidate to the learning data, thereby adding new learning models incrementally.

Apparatus and method for constructing learning data

A phoneme dividing method using a multilevel neural network applied to a phoneme dividing apparatus having a voice input portion, a preprocessor, a multi-layer perceptron (MLP) phoneme dividing portion, and a phoneme border outputting portion includes the steps of: (a) sequentially segmenting and framing voice with digitalized voice samples, extracting characteristic vectors by vocal frames, and extracting an inter-frame characteristic vector of the difference between nearby frames of the characteristic vectors by frames, to thereby normalize the maximum and minimum of the characteristics; (b) storing information on the weight obtained through learning and the standard of the MLP; and (c) reading the weight obtained in the step (b), receiving the characteristic vectors, performing an operation of phoneme border discrimination to generate an output value, discriminating the phoneme border according to the output value, and if the current analyzed frame arrives two frames preceding the final frame of incoming voice, outputting a frame number indicative of the border of phoneme as a final result.

Phoneme dividing method using multilevel neural network

Growing community of inventors

Daejeon, South Korea

Young Jik Lee