Wake-Word Misrecognition Detection for Resource-Limited Audio Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Electronic devices with limited resources face challenges in accurately recognizing wake-up words due to the need for continuous operation, leading to misrecognition issues.
Innovation Solution
An electronic apparatus equipped with a communication interface, memory, and processor that analyzes audio inputs to determine wake-up word recognition, utilizes a neural network model to detect misrecognition, and trains the model based on misrecognized audio clusters to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a high-performance wake up word detection model is mounted in electronic apparatuses with limited resources, then recognition accuracy is improved, but device resource consumption (CPU, memory) increases
Solution Approach 1:
The system segments the wake-up detection process into two distinct stages: a first wake-up detection model running continuously on the electronic apparatus with limited resources, and a second wake-up detection model running on a server with abundant resources. The first model performs initial filtering of audio inputs, and only potentially wake-up-containing segments are transmitted to the server for secondary verification, thereby distributing computational burden and improving overall accuracy without overwhelming the local device resources.
Solution Approach 2:
The first wake-up detection model acts as an intermediary between the audio input and the second wake-up detection model. It pre-processes and filters audio data locally, transmitting only relevant segments to the server, thus reducing the computational load on both the local device and the server while maintaining high recognition accuracy.
2Speed
If the wake up word detection model operates continuously to ensure always-on functionality, then response time is improved, but resource consumption increases
Solution Approach 1:
The system implements periodic action by having the first wake-up detection model continuously monitor audio inputs in real-time on the electronic apparatus, while the second wake-up detection model on the server operates periodically only when triggered by potential wake-up detections from the first model. This approach ensures fast local response while avoiding continuous heavy server resource consumption.
Solution Approach 2:
The detection process is segmented into continuous local monitoring by the first model and periodic server-side verification by the second model, allowing the system to maintain always-on functionality with reduced overall resource consumption.
3Use of energy by moving object
If a simple wake up word detection model is used in electronic apparatuses with limited resources, then resource consumption is reduced, but recognition accuracy deteriorates leading to misrecognition
Solution Approach 1:
The system segments the detection task between a lightweight first model for continuous local monitoring and a sophisticated second model for periodic server-side verification, achieving both low resource consumption and high accuracy through distributed processing.
Solution Approach 2:
The system implements feedback by transmitting results from the first wake-up detection model to the second model for verification. When the first model detects a potential wake-up, it triggers the second model to perform secondary detection, and the final determination is made based on the combined results, thereby improving accuracy while maintaining resource efficiency.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An electronic apparatus is provided. The electronic apparatus includes a communication interface with communication circuitry, a memory configured to store at least one instruction and a processor, and the processor is configured to receive a first audio recognized as a wake up word by an external device from the external device, determine whether the first audio corresponds to the wake up word by analyzing the first audio, based on determining that the first audio does not correspond to the wake up word, obtain a neural network model for detecting a wake up word misrecognition based on the first audio, and transmit information regarding the neural network model to the external device.