Clean Speech Parameter Estimation from Noisy Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speaker and speech recognition systems face performance degradation due to the use of noisy training speech, which results in the creation of spurious models, and current methods for estimating clean speech parameters are computationally complex and inefficient.
Innovation Solution
A method and system that acquires speech signals, estimates noise, computes speech features, and estimates clean model parameters using a processor-based approach, involving modules for noise estimation, feature extraction, and clean parameter estimation, employing techniques like Reverse Psychoacoustic Compensation to derive clean speech parameters from noisy inputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If clean speech vectors are estimated from noisy speech, then recognition accuracy is improved, but computational complexity increases significantly
Solution Approach 1:
The patent extracts only the essential model parameters from the full speech vectors, focusing computation on the critical features that drive recognition accuracy. By separating the essential parameters from the complete vector representation, the system achieves accuracy improvement with reduced computational burden.
Solution Approach 2:
The patent transforms the problem from estimating complete speech vectors to estimating a reduced set of model parameters. This parameter transformation changes the computational landscape, making the clean speech estimation feasible with lower complexity while maintaining the essential information needed for accurate recognition.
2Productivity
If noisy training speech is used for model training, then training efficiency is maintained, but model quality degrades due to spurious models
Solution Approach 1:
The patent converts the harmful effect of noisy training data into a benefit by explicitly modeling and removing the noise characteristics. Instead of treating noise as a problem to be avoided, the system uses noise estimation and removal techniques to transform noisy training speech into clean speech parameters, thereby improving model quality while maintaining training efficiency.
Solution Approach 2:
The patent performs preliminary noise estimation and removal before the actual model training process. By pre-processing the noisy training speech to extract clean speech parameters in advance, the system ensures that the training models are built on clean data without compromising training efficiency, thus preventing spurious model formation.
Data Source
AI summary
A method and system is provided for estimating clean speech parameters from noisy speech parameters. The method is performed by acquiring speech signals, estimating noise from the acquired speech signals, computing speech features from the acquired speech signals, estimating model parameters from the computed speech features and estimating clean parameters from the estimated noise and the estimated model parameters.