Neural Network Speech Signal Isolation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing speech processing systems fail to effectively isolate and reconstruct speech signals from background noise in noisy environments, as frequency components of the speech signal are often masked by background frequencies, leading to degraded signal quality.

Innovation Solution

A neural network-based system that analyzes noisy speech signals to isolate and reconstruct clean speech signals by estimating noise levels across frequency subbands and using a signal blending component to reconstruct the speech signal, reducing or eliminating background noise.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If traditional speech processing systems are used in noisy environments, then the system structure remains simple, but the speech signal isolation and reconstruction quality deteriorates due to frequency masking by background noise

Engineering Contradiction:

Improvespeech signal isolation qualityVSAvoidsystem structure

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent introduces a neural network as an intermediary component between the noisy speech input and the output signal. This neural network mediator learns to separate speech from background noise by training on paired clean and noisy speech data, enabling effective speech isolation without requiring complex manual signal processing pipelines

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system transforms the speech signal into different parameter domains (frequency subbands, time-frequency representations) to facilitate noise separation. By representing the signal in transformed domains and applying the neural network, the system can selectively modify parameters to enhance speech while suppressing noise components

Inventive Principle:

Principle #35Parameter changes

2Reliability

If the speech signal is transmitted in noisy environments, then the transmission range is extended, but the frequency components of the speech signal are masked by background noise, degrading signal quality

Engineering Contradiction:

Improvespeech signal qualityVSAvoidbackground noise masking

Core Design Contradiction:

ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The system performs preliminary noise estimation and speech enhancement before final signal reconstruction. By estimating the noise spectrum in advance and using the neural network to predict clean speech components, the system proactively compensates for noise masking effects rather than attempting correction after damage occurs

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system uses feedback mechanisms where the neural network output is combined with the original noisy signal through a blending component. This feedback loop allows the system to iteratively refine the enhanced speech signal, balancing between noise suppression and preservation of original speech characteristics

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS7620546B2Isolating speech signals utilizing neural networks

Publication Date: 2009.11.17 BLACKBERRY LTD

AI summary

A speech signal isolation system configured to isolate and reconstruct a speech signal transmitted in an environment in which frequency components of the speech signal are masked by background noise. The speech signal isolation system obtains a noisy speech signal from an audio source. The noisy speech signal may then be fed through a neural network that has been trained to isolate and reconstruct a clean speech signal from against background noise. Once the noisy speech signal has been fed through the neural network, the speech signal isolation system generates an estimated speech signal with substantially reduced noise.