Neural Network Speaker Recognition in Noisy Audio Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Connected loudspeakers struggle to recognize command keywords in noisy sound signals, particularly when the speaker has a unique accent or uses an unrepresented language, leading to ineffective command triggering.
Innovation Solution
A method that utilizes an artificial neural network trained on a database of noiseless sound signals from each speaker, combined with environmental noise, to generate sound signatures for identifying speakers and recognizing command keywords, allowing for personalized keyword recognition without phoneme-translation or language comprehension steps, and updates the database to adapt to changing noise environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional on-line machine learning algorithms are used for keyword recognition, then the system can recognize predefined keywords, but it fails to recognize keywords when speakers have unique accents or use unrepresented languages
Solution Approach 1:
The system performs offline training in advance to create speaker-specific sound signatures and language models. During operation, the pre-trained model quickly matches incoming speech against stored speaker profiles, enabling immediate adaptation to new speakers and languages without requiring real-time retraining or extensive computational resources.
Solution Approach 2:
The system creates simplified representations (sound signatures) of speaker characteristics and language patterns from training data. These copied features are stored and reused during recognition, allowing the system to generalize across different speakers and languages while maintaining high recognition accuracy without storing entire language corpora.
2Measurement precision
If speaker-specific training data is collected for each speaker, then recognition accuracy improves, but the data collection and processing time increases
Solution Approach 1:
The system extracts only the essential acoustic features and language patterns from speaker training data, creating compact sound signatures. This extraction process isolates the most discriminative characteristics needed for recognition while discarding redundant information, enabling fast training and recognition with minimal data processing time.
Solution Approach 2:
The system transforms raw speech data into transformed acoustic parameters (sound signatures) that capture speaker-specific characteristics in a condensed form. This parameter transformation reduces the dimensionality and complexity of training data while preserving the essential information needed for accurate speaker and keyword recognition.
3Adaptability or versatility
If phoneme-translation and language comprehension steps are added to handle different languages, then language coverage improves, but system complexity increases
Solution Approach 1:
The system uses a universal sound signature extraction approach that works across multiple languages without requiring language-specific processing pipelines. The same acoustic feature extraction and matching mechanisms handle different languages, eliminating the need for separate phoneme-translation or language comprehension modules while maintaining broad language coverage.
Data Source
AI summary
A method for analysing a noisy sound signal for the recognition of at least one group of control keywords and of a speaker of the analysed noisy sound signal, the noisy sound signal being recorded by a microphone and the method including: supervised training of an artificial neural network using a training database in order to obtain a trained artificial neural network capable of providing, based on a sound signature obtained from a noisy sound signal, a prediction of the speaker and at least one prediction of a group of control keywords, the training database including a plurality of sound signatures, each associated with a speaker and with at least one group of control keywords; calculating a sound signature of the analysed noisy sound signal; using the trained artificial neural network on the calculated sound signature in order to obtain a prediction of the speaker and at least one prediction of a group of control keywords.