DEVICE AND METHOD FOR SPEECH SEPARATION USING SPEAKER EMBEDDING FROM PRELIMINARY SEPERATION
The present invention relates to speech separation technology and more specifically, relates to a device and method for speech separation based on a speaker that divide a repeated block of a speech separation network into a first half part and a second half part and extracts speaker pieces of inform...
Gespeichert in:
Hauptverfasser: | , |
---|---|
Format: | Patent |
Sprache: | eng ; kor |
Schlagworte: | |
Online-Zugang: | Volltext bestellen |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | The present invention relates to speech separation technology and more specifically, relates to a device and method for speech separation based on a speaker that divide a repeated block of a speech separation network into a first half part and a second half part and extracts speaker pieces of information included in a mixed speech signal from a middle separation signal derived from the first half part. According to one embodiment of the present invention, accurate speaker recognition and speech separation can be performed by providing the middle separation signal extracted by the first half part of a speech separation network to the second half part of the speech separation network. The device comprises: a speech input part; a speech separation part; and a speaker information extraction part.
본 발명은 음성 분리 기술에 관한 것으로, 더욱 상세하게는 음성 분리 네트워크의 반복된 블록을 전반부와 후반부로 나누어 전반부에서 도출된 중간 분리 신호로부터 혼합 음성 신호에 포함된 화자 정보들을 추출하는 화자 기반 음성 분리 장치 및 방법에 대한 것이다. 본 발명의 일 실시 예에 따르면, 음성 분리 네트워크의 전반부가 추출한 중간 분리 신호를 음성 분리 네트워크 후반부에 제공하여 정확한 화자 인식 및 음성 분리를 수행할 수 있다. |
---|