公开(公告)号：US20140343935A1

公开(公告)日：2014-11-20

申请号：US14277241

申请日：2014-05-14

Applicant: ELECTRONICS AND TELECOMMUNICATIONS RESEARCH INSTITUTE

Inventor： Ho-Young JUNG ,  Ki-Young PARK ,  Jeom-Ja KANG ,  Yun-Keun LEE

IPC: G10L15/20

CPC classification number: G10L15/20 ,  G10L15/01 ,  G10L15/08 ,  H04R3/005

Abstract: An apparatus and method for performing asynchronous speech recognition using multiple microphones are disclosed. The apparatus includes a microphone selection unit, a signal-to-noise ratio measurement unit, a speech recognition and verification unit, and a final recognition result output unit. The microphone selection unit selects two or more microphones responsive to a user's voice from among a plurality of microphones distributed around the user. The signal-to-noise ratio measurement unit measures the signal to noise ratios of inputs of the selected two or more microphones. The speech recognition and verification unit performs speech recognition using the input of the microphone having a highest signal to noise ratio, and verifies the speech recognition using the inputs of the remaining microphones. The final recognition result output unit outputs the final recognition results of the user's voice based on the results of the speech recognition and verification unit.

Abstract translation: 公开了一种使用多个麦克风执行异步语音识别的装置和方法。 该装置包括麦克风选择单元，信噪比测量单元，语音识别和验证单元以及最终识别结果输出单元。 麦克风选择单元从用户周围分布的多个麦克风中选择响应于用户的语音的两个或更多个麦克风。 信噪比测量单元测量所选择的两个或多个麦克风的输入的信噪比。 语音识别和验证单元使用具有最高信噪比的麦克风的输入来执行语音识别，并且使用剩余麦克风的输入来验证语音识别。 最终识别结果输出单元基于语音识别和验证单元的结果输出用户语音的最终识别结果。

公开(公告)号：US20230134942A1

公开(公告)日：2023-05-04

申请号：US17961830

申请日：2022-10-07

Applicant: ELECTRONICS AND TELECOMMUNICATIONS RESEARCH INSTITUTE

Inventor： Hoon CHUNG ,  Byung-Ok KANG ,  Jeom-Ja KANG ,  Yun-Kyung LEE ,  Hyung-Bae JEON

IPC: G10L15/06 ,  G10L15/187

Abstract: Disclosed herein are an apparatus and method for self-supervised training of an end-to-end speech recognition model. The apparatus includes memory in which at least one program is recorded and a processor for executing the program. The program trains an end-to-end speech recognition model, including an encoder and a decoder, using untranscribed speech data. The program may add predetermined noise to the input signal of the end-to-end speech recognition model, and may calculate loss by reflecting a predetermined constraint based on the output of the encoder of the end-to-end speech recognition model.