Dual Mode Speech Recognition With Dynamic Timeout

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing speech recognition systems face challenges in balancing accuracy and speed, often requiring trade-offs between responding quickly with potentially low-quality results and waiting for higher-quality outputs, especially in environments with variable connectivity and resource constraints.

Innovation Solution

Implementing a dual mode speech recognition system that uses multiple recognizers, where the first received result is accepted if of sufficient quality, and otherwise waits for a second result if the initial result is not, with timeout durations varying based on the quality score to optimize latency and accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the system waits for higher-quality recognition results, then accuracy is improved, but response time increases

Engineering Contradiction:
Improverecognition accuracyVSAvoidresponse time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system dynamically adjusts its behavior based on the quality score of recognition results. When the quality score is high, the system responds immediately; when the quality score is low, the system waits for a second result. This dynamic adaptation allows the system to optimize between speed and accuracy on a per-request basis, resolving the contradiction between fast response and high accuracy.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system changes the timeout duration parameter based on the quality score of the first recognition result. By varying the timeout parameter according to the quality score, the system can wait longer for improved results when the first result is poor, while responding quickly when the first result is sufficient, thus balancing accuracy and response time.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If the system uses multiple recognizers with different qualities, then overall accuracy is improved, but system complexity increases

Engineering Contradiction:
Improverecognition accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system segments the recognition function into multiple independent recognizers with different characteristics (e.g., speed vs. accuracy). Each recognizer is a separate component that can be evaluated independently, and the system selectively uses them based on quality scores. This segmentation allows the system to achieve high accuracy through multiple perspectives while keeping each individual recognizer relatively simple.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system uses quality scores as feedback to determine which recognizer's result to accept. By evaluating the quality of results from multiple recognizers and using this feedback to make decisions about which result to use, the system can maintain high accuracy without needing to manage complex integration logic for all possible combinations of recognizers.

Inventive Principle:
Principle #23Feedback

3Speed

If the system immediately accepts first results, then response speed is improved, but result quality may deteriorate

Engineering Contradiction:
Improveresponse speedVSAvoidrecognition accuracy
Core Design Contradiction:
SpeedVSMeasurement precision

Solution Approach 1:

The system performs a preliminary evaluation of the first recognition result's quality score before making a final decision. This preliminary action allows the system to quickly identify when the first result is sufficient and respond immediately, while only waiting for a second result when the preliminary quality assessment indicates the first result is inadequate, thus maintaining both speed and accuracy.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3413305B1Dual mode speech recognition
Publication Date: 2023.09.06 SOUNDHOUND INC
  • EP3413305B1 patent drawingFigure 1
  • EP3413305B1 patent drawingFigure 2A~2B
  • EP3413305B1 patent drawingFigure 3

AI summary

A dual mode speech recognition system sends speech to two or more speech recognizers. If a first recognition result is received, whose recognition score exceeds a high threshold, the first result is selected without waiting for another result. If the score is below a low threshold, the first result is ignored. At intermediate values of recognition scores, a timeout duration is dynamically determined as a function of the recognition score. The timeout duration determines how long the system will wait for another result. Many functions of the recognition score are possible, but timeout durations generally decrease as scores increase. When receiving a second recognition score before the timeout occurs, a comparison based on recognition scores determines whether the first result or the second result is the basis for creating a response.