METHOD AND APPARATUS FOR RECOGNIZING KEYWORD USING MULTI KEYWORD ENGINE ADAPTIVE TO NOISY ENVIRONMENT

• Main Authors: KIM JONG YUP, JUNG JAE HOON, LEE HYOWON, KIM HEE KYUNG, MOON KYUNG GI
• Format: Patent
• Language: English; Korean
• Published: 11.11.2022
• Subjects: ACOUSTICS; MUSICAL INSTRUMENTS; PHYSICS; SPEECH ANALYSIS OR SYNTHESIS; SPEECH OR AUDIO CODING OR DECODING; SPEECH OR VOICE PROCESSING; SPEECH RECOGNITION

As an operating method of an AI terminal, the operating method comprises: a step of simultaneously inputting an input-received speech signal to a plurality of call word engines and performing a call word recognition through each of the call word engine; a step of selecting a specific call word recognition engine to recognize a call word included in the speech signal based on a call word recognition probability for a result of each of the call word engine; and a step of determining whether or not the call word is included in the speech signal through the specific call word recognition engine, wherein the call word engines are realized with a different call word recognition model. Therefore, the present invention is capable of enabling a high call word recognition rate to be secured. AI 단말의 동작 방법으로서, 입력받은 음성 신호를 동시에 다수개의 호출어 엔진들에 입력하여 각 호출어 엔진을 통해 호출어 인식을 수행하는 단계, 각 호출어 엔진의 결과에 대한 호출어 인식 확률을 기초로 음성 신호에 포함된 호출어를 인식할 특정 호출어 인식 엔진을 선택하는 단계, 그리고 특정 호출어 인식 엔진을 통해 음성 신호에 호출어의 포함 여부를 판단하는 단계를 포함하고, 호출어 엔진들은, 서로 다른 호출어 인식 모델로 구현된다.