Personalized Speech Recognition Model Using Group Scale Matrices

Solution Overview

Problem

Current speech recognition models lack personalization, leading to suboptimal performance for individual users due to generalization across diverse language groups, resulting in reduced accuracy and efficiency.

Innovation Solution

A processor-implemented method that identifies a user's language group and generates a personalized speech recognition model by applying group scale matrices to the model's layers, incorporating user-specific and group-specific scale matrices for improved recognition.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If a general speech recognition model is used, then the model can serve diverse language groups, but the recognition accuracy decreases for individual users

Engineering Contradiction:

Improvemodel adaptabilityVSAvoidrecognition accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent applies local quality by introducing user-specific scale matrices that are uniquely tailored to each user's speech characteristics. Instead of using a single universal model for all users, the system creates personalized model variations by applying individual scale matrices to the shared backbone network, allowing each user to benefit from customized parameter scaling while maintaining the benefits of the shared architecture.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent segments the speech recognition model into a shared backbone network and user-specific scale matrices. This segmentation allows the majority of the model parameters to remain shared across all users (maintaining versatility), while only the scale matrix parameters are personalized for each user (improving accuracy). This divides the model into universal and individual components.

Inventive Principle:

Principle #1Segmentation

2Measurement precision

If a personalized speech recognition model is generated for each user, then recognition accuracy improves, but device complexity increases

Engineering Contradiction:

Improverecognition accuracyVSAvoidmodel complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies local quality by introducing user-specific scale matrices that are uniquely tailored to each user's speech characteristics. Instead of using a single universal model for all users, the system creates personalized model variations by applying individual scale matrices to the shared backbone network, allowing each user to benefit from customized parameter scaling while maintaining the benefits of the shared architecture.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent implements universality through a shared backbone network that serves all users. The scale matrices are applied to this universal model, allowing one base model to serve multiple users with different speech characteristics. This multi-functional approach reduces overall system complexity compared to creating entirely separate models for each user.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Measurement precision

If scale matrices are applied to all layers of the speech recognition model, then personalization effectiveness increases, but processing time increases

Engineering Contradiction:

Improverecognition accuracyVSAvoidprocessing time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies partial action by selectively applying scale matrices to only specific layers of the speech recognition model rather than all layers. The scale matrices are applied to layers where they provide the most benefit for personalization, while skipping layers where the impact is minimal or where computation would be excessive, thus balancing accuracy improvement with processing time constraints.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS10957308B2Device and method to personalize speech recognition model

Publication Date: 2021.03.23 SAMSUNG ELECTRONICS CO LTD

AI summary

Provided is a method and device to personalize a speech recognition model, the device that personalizes a speech recognition model by identifying a language group corresponding to a user, and generating a personalized speech recognition model by applying a group scale matrix corresponding to the identified language group to at least a layer of a speech recognition model.