Voice Recognition Signal Separation in Electronic Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing electronic devices with voice recognition functions face challenges in distinguishing user utterances from other sound signals at long distances, especially in environments with multiple peripheral devices, leading to deteriorated voice recognition performance.
Innovation Solution
An electronic device and method that utilize a combination of microphones and signal processing techniques, including blind source separation and signal characteristic analysis, to identify user utterance voices by separating sound signals into distinct sources and applying predefined criteria such as zero-crossing rate, average volume, and speech detection, to isolate the user's voice from mixed sound signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If multi-channel microphone is used for long-distance voice recognition, then voice recognition coverage is improved, but voice recognition performance deteriorates due to mixed sound signals from peripheral devices
Solution Approach 1:
The patent segments the mixed sound signal into multiple independent sound source signals using blind source separation technology. By dividing the composite signal captured by multi-channel microphones into separate source signals (user voice, TV audio, other devices), the system can selectively process only the user's voice for recognition, thereby maintaining high recognition performance while preserving wide coverage capability
Solution Approach 2:
The patent introduces an intermediary processing layer between signal capture and voice recognition. This intermediary layer performs blind source separation and signal characteristic analysis to identify and extract the user's voice from mixed signals. The intermediary processing enables the system to distinguish user utterances from peripheral device outputs, resolving the contradiction between wide coverage and accurate recognition
2Adaptability or versatility
If speaker recognition technology is applied to classify sound signals, then signal classification capability is improved, but computation resources increase and user information must be stored
Solution Approach 1:
The patent replaces complex speaker recognition technology with simpler, disposable-like signal processing methods. Instead of building and maintaining complex speaker profiles that require substantial computation and data storage, the system uses lightweight blind source separation and real-time signal characteristic analysis (zero-crossing rate, energy, spectral features) to classify and identify sound sources. This approach achieves effective signal classification with minimal computational overhead and without storing user information
Solution Approach 2:
The patent substitutes the mechanical/computational complexity of speaker recognition systems with a more efficient signal processing approach. Rather than using complex machine learning models for speaker identification, the system employs mathematical signal processing techniques (blind source separation algorithms, spectral analysis) that are computationally lighter and do not require storing personal user data, thereby reducing device complexity while maintaining classification capability
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Disclosed is an electronic device capable of improving voice recognition. The electronic device includes a sound receiver, and a processor configured to: acquire a sound signal received by the sound receiver, separate the acquired sound signal into a plurality of sound source signals, detect signal characteristics of each of the plurality of separated sound source signals, and identify a sound source signal corresponding to a user utterance voice among the plurality of sound source signals based on predefined information on a correlation between the detected signal characteristics and the user utterance voice.