Yong F Zhao

US Patent:

7836449, Nov 16, 2010

Filed:

Apr 29, 2005

Appl. No.:

11/117805

Inventors:

Ferhan Elvanoglu - Redmond WA, US
Jin Feng - Redmond WA, US
Nirav A. Kamdar - Redmond WA, US
Yong Zhao - Kirkland WA, US

Assignee:

Microsoft Corporation - Redmond WA

International Classification:

G06F 9/46
G09G 5/00

US Classification:

718100, 718101, 345619

Abstract:

A task infrastructure that allows for the decoupling of the task launch code from the task client code and from the task execution code is provided. The task infrastructure allows the relationship between the task display and launch logic and the task client logic to be defined in a declarative manner, allowing for the segregation of the task display and launch logic from the task client logic. The task infrastructure comprises a manifest that contains a specification of the relationship between the task display and launch logic and the task client logic. The manifest may also define which task clients expose which tasks, the context under which each task is displayed, the arguments that the task needs in order to start execution, the task description to display, and the method in which the task can be launched. At run-time, a task client can read the manifest and decide which task or tasks to expose to an end user under a current context.

US Patent:

20200334538, Oct 22, 2020

Filed:

May 13, 2019

Appl. No.:

16/410741

Inventors:

- Redmond WA, US
Jinyu LI - Redmond WA, US
Yong ZHAO - Redmond WA, US
Yifan GONG - Sammamish WA, US

International Classification:

G06N 3/08
G10L 15/16
G06N 3/04
G10L 15/183

Abstract:

Embodiments are associated with conditional teacher-student model training. A trained teacher model configured to perform a task may be accessed and an untrained student model may be created. A model training platform may provide training data labeled with ground truths to the teacher model to produce teacher posteriors representing the training data. When it is determined that a teacher posterior matches the associated ground truth label, the platform may conditionally use the teacher posterior to train the student model. When it is determined that a teacher posterior does not match the associated ground truth label, the platform may conditionally use the ground truth label to train the student model. The models might be associated with, for example, automatic speech recognition (e.g., in connection with domain adaptation and/or speaker adaptation).

US Patent:

20200335122, Oct 22, 2020

Filed:

Jun 7, 2019

Appl. No.:

16/434665

Inventors:

- Redmond WA, US
Yong ZHAO - Redmond WA, US
Jinyu LI - Redmond WA, US
Yifan GONG - Sammamish WA, US

International Classification:

G10L 25/03
G10L 25/30
H04R 29/00
H04R 5/04
G06F 17/18
G06N 3/08
G06N 20/00

Abstract:

To generate substantially condition-invariant and speaker-discriminative features, embodiments are associated with a feature extractor capable of extracting features from speech frames based on first parameters, a speaker classifier capable of identifying a speaker based on the features and on second parameters, and a condition classifier capable of identifying a noise condition based on the features and on third parameters. The first parameters of the feature extractor and the second parameters of the speaker classifier are trained to minimize a speaker classification loss, the first parameters of the feature extractor are further trained to maximize a condition classification loss, and the third parameters of the condition classifier are trained to minimize the condition classification loss.