Personalized Speech Signal Filtering for Competing Speaker Noise
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video conferencing systems struggle to effectively reduce background noise, particularly when competing speech is at a similar level to the target speaker, often failing to distinguish and attenuate speech from individuals close to the target user.
Innovation Solution
A personalized machine learning-based system that trains on the user's speech to attenuate competing individual speech and background noise, utilizing a neural network that adapts to the user's environment and noise patterns, including ambient sounds and reverberation, and maintains a database of speech features for rapid personalization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If traditional noise reduction subsystems are used, then background ambient and transient noises can be reduced to some extent, but the system becomes ineffective when competing speech is approximately the same level as target speaker speech
Solution Approach 1:
The system changes the parameter of noise reduction from a fixed approach to a dynamic, adaptive approach. The machine learning model continuously learns from audio data to adjust its noise reduction parameters in real-time, allowing it to distinguish between background noise and competing speech based on temporal patterns and acoustic characteristics specific to each user's environment.
Solution Approach 2:
The system implements self-service by automatically adapting to each user's specific acoustic environment without manual configuration. The machine learning model trains on audio data collected during communication sessions, enabling the system to autonomously learn and attenuate noise patterns specific to each user's location and surroundings.
2Reliability
If a personalized machine learning model is trained for each user, then noise attenuation effectiveness improves, but system complexity and training requirements increase
Solution Approach 1:
The system achieves universality by using a single machine learning model architecture that can serve multiple users. The model is designed to be trained on audio data from any user and then applied to attenuate noise for that user, allowing the same system framework to handle diverse acoustic environments and user requirements without requiring separate specialized systems for each user.
Solution Approach 2:
The system performs preliminary action by collecting and training on audio data during initial communication sessions before actual use. This training phase occurs in advance, allowing the machine learning model to learn and store noise attenuation parameters specific to each user's environment, so that when real-time communication occurs, the pre-trained model can immediately provide effective noise reduction.
3Measurement precision
If real-time noise attenuation is applied, then speech clarity improves, but processing time and computational resources increase
Solution Approach 1:
The system performs preliminary action by pre-training the machine learning model on audio data collected during communication sessions before real-time processing is needed. This advance training allows the model to internalize noise attenuation patterns, so that during real-time operation, the model can quickly apply learned transformations without requiring extensive computational processing time.
Solution Approach 2:
The system replaces traditional mechanical signal processing approaches with machine learning-based transformations. Instead of using fixed algorithms that require complex real-time calculations, the system uses a trained neural network that has already learned optimal transformation parameters, enabling faster and more efficient real-time noise attenuation while maintaining high speech clarity.
Data Source
AI summary
In one example embodiment, speech signals are received from a user during a communication session. The received speech signals contain noise including speech of other individuals. The received speech signals are transformed by a machine learning model to produce transformed speech signals corresponding to the received speech signals with a reduced amount of the noise. The machine learning model is trained with speech of the user satisfying a noise threshold and collected during one or more communication sessions.