Modular Arbitration for Extensible Speech Recognition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional multi-class classifiers in speech recognition UI systems are not extensible for independently developed applications, limiting their ability to handle new or unknown applications effectively.

Innovation Solution

A modular UI system architecture where each module implements a subset of the UI functionality, allowing independent development and arbitration by a single system that analyzes user requests and prioritizes relevant modules based on probability outputs from classifiers trained using a shared reference dataset, enabling extensibility without modifying existing modules or the arbitration system.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a multi-class classifier is used for arbitration in speech recognition UI systems, then arbitration performance for known applications is improved, but extensibility for independently developed applications deteriorates

Engineering Contradiction:
Improvearbitration performanceVSAvoidextensibility
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The system segments the arbitration function into independent module-specific classifiers rather than a single multi-class classifier. Each module maintains its own classifier that independently evaluates user inputs, allowing modules to be developed and updated independently while maintaining arbitration performance through modular, interchangeable components.

Inventive Principle:
Principle #1Segmentation

2Extent of automation

If a single arbitration system processes inputs from multiple applications, then centralized control is improved, but system complexity deteriorates

Engineering Contradiction:
Improvecentralized controlVSAvoidsystem complexity
Core Design Contradiction:
Extent of automationVSDevice complexity

Solution Approach 1:

Each module performs self-service by maintaining its own classifier that independently evaluates user inputs and determines its own relevance. This eliminates the need for complex centralized arbitration logic, as each module autonomously assesses its suitability for handling the input, thereby maintaining centralized coordination while reducing overall system complexity.

Inventive Principle:
Principle #25Self-service

3Productivity

If modules are developed independently, then development efficiency is improved, but integration complexity deteriorates

Engineering Contradiction:
Improvedevelopment efficiencyVSAvoidintegration complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The system implements a universal arbitration framework where all modules interface through a common probability output mechanism. Each module's classifier produces a standardized probability indicating input relevance, allowing independently developed modules to be integrated without modification. This universal interface standard enables seamless integration while preserving independent development efficiency.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP3201770B1Methods and apparatus for module arbitration
Publication Date: 2020.06.03 NUANCE COMMUNICATIONS INC
  • EP3201770B1 patent drawingFigure 1
  • EP3201770B1 patent drawingFigure 2
  • EP3201770B1 patent drawingFigure 3

AI summary

Methods and apparatus to process a user input on independent applications that provide classifier outputs to an arbitration module, which selects one of the application to respond to the user input. The classifier outputs include a probability that the user input is in domain for the application functionality.