Mobile Speech Enhancement Using Compact Spectra Preselection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional noise reduction methods in mobile communication devices are limited by high computational load and energy consumption, making them unsuitable for devices with limited resources, especially in noisy environments where time-varying non-stationary noise is prevalent.
Innovation Solution
A method that involves acquiring distorted signal frames, computing orthogonal transforms, and using compact spectra prototypes and noise models to remove noise, with preselection and a priori knowledge models to reduce computational effort and enhance speech signal recovery, employing dictionaries of speech and noise gains, and covariance matrices to estimate noise and speech spectra efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a priori knowledge model with full dimension spectra is used for noise reduction, then speech enhancement quality is improved, but computational load and energy consumption increase significantly
Solution Approach 1:
The patent segments the computational process into two distinct stages: preselection using compact spectra of first dimension (reduced dimensionality) and precise probability evaluation using compact spectra of second dimension (higher dimensionality). This segmentation allows the system to benefit from both low computational complexity and high accuracy without requiring full-dimension processing throughout the entire pipeline.
Solution Approach 2:
The patent performs preliminary preselection of composite states using compact spectra of first dimension before conducting the computationally intensive probability evaluation. This preliminary action filters out unlikely states early, reducing the number of states that require full probability computation and thereby significantly lowering overall computational load and energy consumption.
2Measurement precision
If a priori knowledge model with full dimension spectra is used for noise reduction, then speech enhancement quality is improved, but computational complexity increases
Solution Approach 1:
The computational complexity is segmented across two processing stages with different dimensional requirements. The preselection stage operates in reduced dimensionality space (first dimension compact spectra), while the refinement stage operates in higher dimensionality space (second dimension compact spectra). This segmentation prevents the need for full-dimension processing throughout the entire algorithm.
Solution Approach 2:
The preselection step performs preliminary filtering using low-dimensional compact spectra before the main probability evaluation step. This preliminary action reduces the search space from all possible composite states to a manageable subset, significantly simplifying the subsequent computational task and reducing overall algorithmic complexity.
3Use of energy by moving object
If conventional filtration techniques are used, then computational load is reduced, but effectiveness in non-stationary noise is severely limited
Solution Approach 1:
The patent implements a dynamic two-stage processing approach where the system adapts its computational strategy based on the input signal characteristics. The preselection stage dynamically identifies promising composite states, and the second stage dynamically evaluates probabilities only for those selected states. This dynamic adaptation maintains high effectiveness in non-stationary noise while controlling computational load.
Solution Approach 2:
The compact spectra of first dimension serve as an intermediary representation between the raw distorted signal and the final enhanced output. This intermediate representation enables efficient preselection and state filtering, bridging the gap between simple filtration and complex full-dimension spectral analysis, thereby achieving both efficiency and effectiveness.
4Measurement precision
If a priori knowledge model is used in mobile devices, then speech enhancement is improved, but battery life decreases due to fast exhaustion
Solution Approach 1:
The preselection step performs preliminary filtering using computationally efficient compact spectra before the main probability evaluation. This preliminary action significantly reduces the number of states requiring intensive computation, thereby reducing overall energy consumption and extending battery life in mobile devices while maintaining speech enhancement quality.
Solution Approach 2:
The processing pipeline is segmented into two stages with different computational demands. The first stage (preselection) consumes minimal energy and identifies candidate states, while the second stage (probability evaluation) consumes more energy but operates on a reduced set of candidates. This segmentation makes the overall system energy-efficient enough for mobile device operation.
Data Source
AI summary
A method of enhancing distorted signal having a speech and noise component, with a processing device having memory with stored training information T, comprising a step of removing noise in spectrum domain according to a noise and speech model to obtain a clear signal spectrum, wherein the training information T comprises dictionaries of compact spectra prototypes of speech and noise, speech gains and noise gains forming together composite states and it comprises probabilities of state history. Dictionaries of compact spectra prototypes comprise compact spectra prototypes of a first dimension P1 and of a second dimension P2 where second dimension P2 is higher or equal to the first dimension P1. A mobile communication device having a battery and processing device with connected memory, radio transceiver and audio input, according to the invention is adapted to receive audio signal from audio input and execute the method according to the invention on this signal and then transmit this signal with radio transceiver. A computer program product according to the invention when executed by mobile communicating device causes execution of the method according to the invention.