Speech synthesis method and apparatus using adversarial learning technique

• Main Authors: LEE MOA, CHANG JOON HYEOK
• Format: Patent
• Language: English; Korean
• Published: 25.08.2023
• Subjects: ACOUSTICS; CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; MUSICAL INSTRUMENTS; PHYSICS; SPEECH ANALYSIS OR SYNTHESIS; SPEECH OR AUDIO CODING OR DECODING; SPEECH OR VOICE PROCESSING; SPEECH RECOGNITION

A speech synthesis method using an adversarial learning technique according to one embodiment comprises: a step of receiving speech data input; a step of learning an adversarial model for synthesizing a speech based on the speech data input; and a step of synthesizing a frame of a target speech using the adversarial model, wherein the step of synthesizing the frame of the speech may comprise synthesizing the frame of the target speech in a non-automatic regression method. Therefore, the present invention is capable of having an advantage of being applicable to a real-time speech synthesis program. 일 실시예에 따른 적대적 학습 기법을 이용한 음성 합성 방법은, 음성 데이터 입력을 수신하는 단계, 상기 음성 데이터 입력에 기반하여 음성을 합성하기 위한 적대적 모델을 학습하는 단계 및 상기 적대적 모델을 이용하여 타겟 음성의 프레임을 합성하는 단계를 포함하고, 상기 음성의 프레임을 합성하는 단계는, 비자동 회귀 방식으로 상기 타겟 음성의 프레임을 합성하는 것을 포함할 수 있다.