Switchable Noise Reduction Profiles for Context-Adaptive Speech Suppression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional noise suppression systems are ineffective in suppressing non-random noise components such as human background sounds, and require separate models for different situations, leading to inefficient and costly noise suppression.
Innovation Solution
A machine learning system, such as a deep neural network, is trained on paired clean and noisy speech signals to adapt noise suppression behavior based on contextual profiles, allowing dynamic adjustment of noise suppression parameters without needing multiple models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional noise suppression systems are used, then stationary noise can be suppressed to some degree, but non-random noise components such as human background sounds cannot be effectively suppressed
Solution Approach 1:
The system dynamically adapts its noise suppression behavior by selecting from multiple pre-trained noise reduction profiles based on contextual information. The machine learning model transitions from a static conventional approach to a dynamic system that adjusts its suppression characteristics in real-time according to the detected context, enabling effective handling of both stationary and non-stationary noise components.
Solution Approach 2:
The system changes the operational parameters of the noise suppression by selecting different pre-trained profiles, each with distinct suppression characteristics. Instead of modifying a single model's parameters, the system switches between multiple trained models (profiles) that have been trained on different noise types, thereby adapting to various noise scenarios including human background sounds.
2Reliability
If separate machine learning models are used for different noise suppression situations, then each situation can be optimized, but storage requirements and computational overhead increase
Solution Approach 1:
The system creates a universal noise suppression solution by training a single machine learning model on multiple noise reduction profiles covering different situations (business calls, family calls, casual calls). This multi-functional model can adapt to various noise scenarios without requiring separate dedicated models for each situation, thereby reducing storage requirements while maintaining optimized performance across different contexts.
Solution Approach 2:
The system merges multiple noise suppression functionalities into a single machine learning model by combining different training profiles during the training phase. Instead of maintaining separate models for business calls, family calls, and casual calls, the system integrates all these functionalities into one model that selects appropriate suppression behavior based on the input context, reducing the quantity of stored models while preserving specialized performance.
3Adaptability or versatility
If multiple pre-trained noise reduction profiles are stored, then different noise situations can be handled, but device complexity and storage costs increase
Solution Approach 1:
The system segments the noise suppression task into distinct pre-trained profiles (business call profile, family call profile, casual call profile), each specialized for a specific noise situation. By dividing the overall noise suppression problem into these manageable segments and organizing them within a single unified model, the system achieves high adaptability while keeping the implementation complexity manageable through structured organization.
4Reliability
If noise suppression parameters are dynamically adjusted, then perceived quality and appropriateness increase, but processing complexity increases
Solution Approach 1:
The system performs preliminary action by pre-training the machine learning model with multiple noise reduction profiles covering different situations before actual noise suppression is needed. This advance preparation allows the model to have all necessary suppression strategies ready, enabling dynamic parameter adjustment during operation without requiring complex real-time calculations or adaptive learning, thus improving perceived quality while limiting processing complexity.
Data Source
AI summary
Disclosed are systems, methods, and other implementations for noise suppression, including a method that includes obtaining a sound signal sample, determining a noise reduction profile, from a plurality of noise reduction profiles, for processing the obtained sound signal sample, and processing the sound signal sample with a machine learning system to produce a noise suppressed signal. The machine learning system implements (executes) a single machine learning model trained to controllably suppress noise in input sound signals according to the plurality of noise reduction profiles. The processing of the sound signal sample is performed according to the determined noise reduction profile.