公开(公告)号：US11978464B2

公开(公告)日：2024-05-07

申请号：US17757122

申请日：2021-01-22

Applicant: GOOGLE LLC

Inventor： Willem Bastiaan Kleijn ,  Andrew Storus

IPC: G10L19/00 ,  G10L19/038 ,  G10L19/04 ,  G10L21/02 ,  G06N3/02

CPC classification number: G10L19/038 ,  G10L19/04 ,  G10L21/02 ,  G06N3/02 ,  G10L19/00

Abstract: A method includes receiving sampled audio data corresponding to utterances and training a machine learning (ML) model, using the sampled audio data, to generate a high-fidelity audio stream from a low bitrate input bitstream. The training of the ML model includes de-emphasizing the influence of low-probability distortion events in the sampled audio data on the trained ML model, where the de-emphasizing of the distortion events is achieved by the inclusion of a term in an objective function of the ML model, which term encourages low-variance predictive distributions of a next sample in the sampled audio data, based on previous samples of the audio data.

公开(公告)号：US20230352036A1

公开(公告)日：2023-11-02

申请号：US17757122

申请日：2021-01-22

Applicant: GOOGLE LLC

Inventor： Willem Bastiaan Kleijn ,  Andrew Storus

IPC: G10L19/038 ,  G10L21/02 ,  G10L19/04

CPC classification number: G10L19/038 ,  G10L21/02 ,  G10L19/04

Abstract: A method includes receiving sampled audio data corresponding to utterances and training a machine learning (ML) model, using the sampled audio data, to generate a high-fidelity audio stream from a low bitrate input bitstream. The training of the ML model includes de-emphasizing the influence of low-probability distortion events in the sampled audio data on the trained ML model, where the de-emphasizing of the distortion events is achieved by the inclusion of a term in an objective function of the ML model, which term encourages low-variance predictive distributions of a next sample in the sampled audio data, based on previous samples of the audio data.