Voice Command Correction Using Confidence-Guided Error Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional speech recognition systems struggle to accurately process voice commands in noisy environments, differentiate between voices, and handle accents and dialects, often failing to distinguish between foreground and background noises, especially during pauses in speech.

Innovation Solution

A method for correcting voice commands during runtime by recognizing errors and adapting processing based on confidence information, including domain and entity determination, error separation, classification, and correction, with specialized error correction components to enhance recognition accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional error correction methods process the entire audio through the algorithm, then all possible errors can be checked, but processing time and computational resources are wasted when there is no background noise

Engineering Contradiction:
Improveerror detection accuracyVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary analysis of the audio to detect background noise before executing the full error correction algorithm. Confidence information is generated during speech recognition to pre-identify segments that may contain errors, allowing the system to skip unnecessary processing of clean audio segments and focus computational resources only on problematic areas.

Inventive Principle:
Principle #10Preliminary action

2Reliability

If the algorithm processes audio during pauses in speech, then errors can be detected, but it becomes difficult to distinguish between foreground and background noises

Engineering Contradiction:
Improveerror detection accuracyVSAvoidnoise differentiation complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system introduces confidence information as an intermediary metric that quantifies the likelihood of errors in speech recognition results. This confidence score, generated during the speech recognition process, serves as a mediator to identify segments requiring error correction without requiring complex analysis of background versus foreground noise during pauses.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If background noise extraction methods are used, then noise can be removed from the audio, but the methods have limitations in detecting background noise as it strongly depends on the use case, environment and context

Engineering Contradiction:
Improvenoise removal effectivenessVSAvoidenvironmental adaptability
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The system uses confidence information from speech recognition as feedback to identify segments that may contain background noise or errors. This feedback mechanism allows the system to adapt to different environments and use cases dynamically, focusing error correction efforts on segments where the speech recognition system expresses uncertainty, rather than relying on pre-configured noise profiles for specific environments.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4571567B1Method, computer program, apparatuses, and speech processing system for correcting a voice command
Publication Date: 2026.05.13 DEUTSCHE TELEKOM AG
  • EP4571567B1 patent drawingFigure 1
  • EP4571567B1 patent drawingFigure 2
  • EP4571567B1 patent drawingFigure 2

AI summary

Examples relate to a method, a computer program, an apparatus and a speech processing system for correcting a voice command. The method (10) comprises obtaining (12) an audio representation of the voice command and speech recognizing (14) the voice command using the audio representation to obtain a text representation of the voice command and recognition confidence information indicating a confidence on whether the text representation matches the audio representation. The method (10) further comprises, based on the recognition confidence information, error processing (16) the audio representation and the text representation or non-error processing (18) the audio representation.