Neural Network Speech Denoising via Silent Interval Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing speech denoising technologies face challenges in effectively removing nonstationary noise from audio recordings, as they often require manual specification of silent intervals and are inadequate for time-varying noise profiles.

Innovation Solution

A deep neural network architecture that integrates silent interval detection, noise estimation, and noise removal components, utilizing LSTM structures and 2D convolutional encoders to automatically identify and utilize silent intervals for robust denoising of speech signals, even in the presence of nonstationary noise.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If manual specification of silent intervals is used, then denoising can be performed, but the system cannot adapt to time-varying noise profiles

Engineering Contradiction:

Improveadaptability to time-varying noiseVSAvoidmanual specification requirement

Core Design Contradiction:

Adaptability or versatilityVSEase of operation

Solution Approach 1:

The system automatically detects silent intervals and estimates noise profiles without requiring manual specification. The neural network autonomously identifies silent periods in the audio signal and uses these to learn time-varying noise characteristics, enabling the system to serve itself rather than requiring human intervention for each denoising task.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system transitions from static manual noise profile specification to dynamic automatic detection. The neural network continuously adapts to changing noise conditions by detecting silent intervals throughout the audio signal and updating noise estimates in real-time, making the system responsive to time-varying noise profiles.

Inventive Principle:

Principle #15Dynamics

2Reliability

If deep neural network with silent interval detection is used, then denoising performance is improved, but device complexity increases

Engineering Contradiction:

Improvedenoising performanceVSAvoidneural network architecture

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The neural network is divided into distinct functional modules: a silent interval detection component that identifies silent periods in the audio signal, and a noise estimation component that uses these detected intervals to estimate noise profiles. This segmentation allows each module to specialize in a specific task, improving overall performance while making the complex system more manageable and interpretable.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS11894012B2Neural-network-based approach for speech denoising

Publication Date: 2024.02.06 SOFTBANK CORPORATION

AI summary

Disclosed are methods, systems, device, and other implementations, including a method that includes receiving an audio signal representation, detecting in the received audio signal representation, using a first learning model, one or more silent intervals with reduced foreground sound levels, determining based on the detected one or more silent intervals an estimated full noise profile corresponding to the audio signal representation, and generating with a second learning model, based on the received audio signal representation and on the determined estimated full noise profile, a resultant audio signal representation with a reduced noise level.