Offline Speech Command Recognition Using Pseudo-Word Rejection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing speech recognition systems in industrial environments face challenges in accurately distinguishing valid commands from invalid ones, particularly in offline settings without internet connectivity, which can lead to operational inefficiencies and potential hazards.

Innovation Solution

A method utilizing a set of pseudo words, each representing multiple real words, is employed to map acoustic signals to phonemes and graphemes, followed by a grammar-based language model to determine valid commands, ensuring fast and safe operation by rejecting commands containing pseudo words.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If speech recognition is performed locally without cloud connection, then cyber security and offline operation are improved, but speech recognition accuracy deteriorates due to inability to distinguish valid commands from invalid ones

Engineering Contradiction:
Improvecyber securityVSAvoidspeech recognition accuracy
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The system pre-generates a set of pseudo words that are phonetically similar to real words before speech recognition occurs. These pseudo words are stored in the local device and used during offline speech recognition to pre-establish discrimination capabilities between valid and invalid commands, enabling accurate command determination without cloud connection.

Inventive Principle:
Principle #10Preliminary action

2Adaptability or versatility

If a large vocabulary is used for speech recognition, then speech recognition coverage is improved, but processing speed deteriorates due to increased computational complexity

Engineering Contradiction:
Improvespeech recognition coverageVSAvoidcommand determination speed
Core Design Contradiction:
Adaptability or versatilityVSSpeed

Solution Approach 1:

The vocabulary is segmented into two distinct sets: real words representing valid commands and pseudo words representing invalid commands. This segmentation allows the system to maintain a comprehensive vocabulary for coverage while enabling rapid discrimination through set-based comparison, significantly improving processing speed for command determination.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Instead of using the entire large vocabulary for both recognition and validation, the system creates a simplified copy in the form of pseudo words that capture the phonetic patterns of invalid commands. This copying approach reduces computational complexity while maintaining comprehensive speech recognition coverage.

Inventive Principle:
Principle #26Copying

3Reliability

If phonetic approximation is used to handle unknown words, then speech recognition robustness is improved, but command validity discrimination deteriorates

Engineering Contradiction:
Improvespeech recognition robustnessVSAvoidcommand validity discrimination
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

Pseudo words serve as an intermediary between acoustic signals and command validation. They act as a mediator that captures phonetic approximations of invalid commands while maintaining distinct boundaries from real words, enabling the system to robustly handle unknown speech patterns while precisely discriminating command validity through set membership checking.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4292083B1Method, device, computer program and computer readable storage medium for determining a command
Publication Date: 2025.08.13 EATON INTELLIGENT POWER LTD
  • EP4292083B1 patent drawingFigure 1
  • EP4292083B1 patent drawingFigure 2
  • EP4292083B1 patent drawingFigure 3~4

AI summary

A method for determining a command is provided, comprising: providing a set of pseudo words, wherein each pseudo word of the set of pseudo words is representative for more than one real word of a set of real words, which is provided for a pronunciation model; mapping an acoustic signal to phonemes by an acoustic model; determining at least one output word dependent on the phonemes and the pronunciation model; and determine the command dependent on the at least one output word, the set of pseudo words and a set of command words, wherein dependent on the command a device performs an action. In addition, a device, a computer program and a computer readable storage medium are provided.