Voice Command Recognition with Local Semantic Matching

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing voice command recognition systems for human-robot communication require users to memorize specific commands, are cumbersome due to intensive training, and rely on computationally demanding Large Language Models (LLMs) that cannot be deployed locally, leading to high latency and resource consumption.

Innovation Solution

Implement a lightweight semantic similarity system using a k-nearest neighbor classifier with multiple reference embeddings and a sentence transformer model to allow for flexible command recognition, enabling local deployment and reducing latency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If Large Language Models (LLMs) are used for voice command recognition, then command understanding accuracy is improved, but computational resource consumption increases and latency increases

Engineering Contradiction:

Improvecommand understanding accuracyVSAvoidcomputational resource consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent creates a simplified copy of the command recognition system using a k-Nearest Neighbor classifier with pre-computed embeddings, replacing the complex LLM while maintaining the ability to accurately recognize commands from natural speech variations

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The system segments the command recognition task into separate components: speech-to-text conversion, embedding generation, and classification, allowing each component to be optimized independently and enabling local deployment without requiring full LLM computational power

Inventive Principle:

Principle #1Segmentation

2Measurement precision

If Large Language Models (LLMs) are used for voice command recognition, then command understanding accuracy is improved, but system latency increases

Engineering Contradiction:

Improvecommand understanding accuracyVSAvoidsystem latency

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system pre-computes and stores embeddings for multiple command variations in advance, so that during actual command recognition, the system only needs to perform fast k-Nearest Neighbor searches rather than generating responses in real-time, significantly reducing latency

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

Inventive Principle:

Principle #26Copying

3Measurement precision

If specific commands are required for robot control, then command recognition accuracy is improved, but ease of operation deteriorates due to intensive user training

Engineering Contradiction:

Improvecommand recognition accuracyVSAvoiduser training requirement

Core Design Contradiction:

Measurement precisionVSEase of operation

Solution Approach 1:

The system is designed to handle multiple ways of expressing the same command through semantic similarity matching, allowing users to interact with the robot using natural language variations without needing to learn specific predetermined commands

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system incorporates feedback loops that continuously improve its understanding by learning from user interactions, allowing it to adapt to different speaking styles and contexts over time, thereby reducing the need for formal user training

Inventive Principle:

Principle #23Feedback

4Adaptability or versatility

If cloud-based LLMs are used for voice command recognition, then command understanding capability is improved, but device complexity and connectivity requirements increase

Engineering Contradiction:

Improvecommand understanding capabilityVSAvoidconnectivity requirements

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent extracts the essential command recognition functionality from cloud-based LLMs and implements it locally using a k-Nearest Neighbor classifier with pre-computed embeddings, eliminating the need for continuous cloud connectivity while maintaining core command understanding capabilities

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The system creates a local copy of the command recognition model that can operate independently without cloud connectivity, preserving the essential functionality while reducing device complexity and connectivity requirements

Inventive Principle:

Principle #26Copying

Data Source

PatentUS20250322831A1Voice command recognition for human-robot communication

Publication Date: 2025.10.16 INTEL CORP

AI summary

Techniques for the use of verbal commands in human-robot communication. The number of tasks the robot can perform is limited to a specific set, while providing syntactic flexibility to users. The system includes two components: a speech recognizer for speech-to-text conversion and a natural language understanding module that maps the text to a command for the robot. After speech is transcribed to text, a nearest neighbor classifier can be applied in the high dimensional space of embedding tokens. Multiple variants of each command are provided in a database of reference embeddings, and the classifier can identify the k nearest reference embedding tokens to determine the command. The text similarity model allows for quick detection solutions to be deployed locally on a robot or other device. Local deployment reduces potential latency caused by a cloud connection, which can be important in many assistant robot applications.