Adaptive Audio Noise Reduction for Multi-Speaker Conferencing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Audio conference systems face challenges in effectively processing audio signals when both a nearby speaker and a loudspeaker are active, as existing noise reduction techniques perform poorly in such scenarios, leading to degraded audio quality.
Innovation Solution
The system employs two noise reduction models trained differently: one for single-speaker scenarios and another for multiple-speaker scenarios, selecting the appropriate model based on the energy level of the loudspeaker output to enhance audio quality by adapting to varying speaking conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single noise reduction model is used for all scenarios, then the device complexity is reduced, but the audio quality deteriorates when multiple speakers are present
Solution Approach 1:
The patent divides the noise reduction task into multiple specialized models: a first model trained for single-speaker scenarios and a second model trained for multiple-speaker scenarios. The system segments the processing based on the number of active speakers detected, routing audio through the appropriate model to optimize performance for each specific condition.
Solution Approach 2:
The system dynamically selects which noise reduction model to apply based on real-time detection of the number of active speakers. This dynamic adaptation allows the system to switch between different processing configurations depending on the acoustic environment, maintaining high audio quality across varying conditions.
2Reliability
If different noise reduction models are trained for different scenarios, then the audio quality is improved, but the device complexity increases
Solution Approach 1:
The system performs preliminary training of multiple noise reduction models during the design phase, with each model specialized for specific speaker scenarios. This preliminary action prepares the system in advance for different conditions, allowing rapid selection during operation without real-time computation overhead.
Solution Approach 2:
The patent changes the training parameters and data composition for different models: the first model is trained primarily on single-speaker audio data, while the second model is trained on multi-speaker audio data. This parameter differentiation allows each model to specialize in its target scenario while sharing the same underlying architecture.
3Object-affected harmful factors
If noise reduction techniques are applied when loudspeaker output is present, then the noise is reduced, but the audio quality deteriorates due to echo and feedback
Solution Approach 1:
The patent applies different noise reduction processing locally based on the acoustic scenario: when a loudspeaker is actively outputting audio, the system detects this condition and selects the appropriate model that accounts for echo and feedback characteristics, applying localized processing tailored to the specific harmful factors present in that environment.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for reducing audio noise are disclosed. In one aspect, a method includes the actions of receiving first audio data of a user utterance. The actions further include determining an energy level of second audio data being outputted by the loudspeaker. The actions further include selecting a model from among (i) a first model that is trained using first audio data samples that each encode speech from one speaker and (ii) a second model that is trained using second audio data samples that each encode speech from either one speaker or two speakers. The actions further include providing the first audio data as an input to the selected model. The actions further include receiving processed first audio data. The actions further include outputting the processed first audio data.