公开(公告)号：US20240289491A1

公开(公告)日：2024-08-29

申请号：US18629401

申请日：2024-04-08

Applicant: SAMSUNG ELECTRONICS CO., LTD.

Inventor： Jisi ZHANG ,  Md Asif JALAL ,  Karthikeyan SARAVANAN ,  Pablo PESO PARADA ,  Mete OZAY

IPC: G06F21/62 ,  G10L15/02 ,  G10L15/06 ,  G10L15/16 ,  G10L15/22 ,  G10L15/30 ,  G10L21/007 ,  G10L21/0208

CPC classification number: G06F21/6254 ,  G10L15/02 ,  G10L15/063 ,  G10L15/16 ,  G10L15/22 ,  G10L15/30 ,  G10L21/007 ,  G10L21/0208

Abstract: Broadly speaking, the present disclosure relates to a computer-implemented method for training a machine learning, ML, automatic speech recognition, ASR, model. The method comprises injecting a speaker anonymiser, which is configured to cause the ML ASR model to generate anonymised acoustic embeddings for the ML ASR model, at one or more layers of the ML ASR model, and suitably training the ML ASR model including the speaker anonymiser on audio data comprising an utterance with one or more words to be recognised. Correspondingly, there is also described a computer implemented method for performing automatic speech recognition using the trained ML ASR model and system for training/inference thereof.

公开(公告)号：US20240013775A1

公开(公告)日：2024-01-11

申请号：US18371233

申请日：2023-09-21

Applicant: SAMSUNG ELECTRONICS CO., LTD.

Inventor： Pablo PESO PARADA ,  Agnieszka DOBROWOLSKA ,  Karthikeyan SARAVANAN ,  Mete OZAY

IPC: G10L15/06 ,  G10L21/0216

CPC classification number: G10L15/063 ,  G10L21/0216

Abstract: A method of obtaining a patched signal for training a model for use in at least one of a speech and an audio recognition is disclosed. The method comprises obtaining a first signal, wherein the first signal is at least one of a speech and an audio signal, modifying the first signal to obtain at least one second signal, dividing the first signal and the at least one second signal respectively into a plurality of first patches and a plurality of second patches, wherein each one of the plurality of first patches comprises a respective part of the first signal and each one of the plurality of second patches comprises a respective part of the at least one second signal and mixing selected ones of the plurality of first patches and the plurality of second patches to obtain a patched signal.