Machine Learning Audio Distortion Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies lack an effective method to quantify and measure distortion in audio signals, particularly for short-duration signals where traditional Total Harmonic Distortion (THD) analysis is not applicable.
Innovation Solution
The development of a system that utilizes a machine learning (ML) model, specifically a convolutional neural network (CNN) or other neural networks like LSTM and Transformer, to estimate the level of distortion in audio signals by converting audio signals into images and applying the trained ML model for distortion estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional THD analysis is used to measure distortion, then measurement precision is improved for long-duration signals, but the method becomes inapplicable for short-duration signals
Solution Approach 1:
The patent changes the fundamental parameters of distortion measurement by transitioning from traditional THD analysis (which requires long-duration signals and frequency domain transformation) to a time-domain machine learning approach. The system uses neural networks trained on short signal segments to directly predict distortion levels, eliminating the duration requirement while maintaining measurement precision through learned patterns in the time domain
Solution Approach 2:
The patent substitutes the mechanical/mathematical THD analysis system with a machine learning system. Instead of using Fourier transforms and harmonic analysis, the invention employs trained neural networks (CNN, RNN, LSTM, Transformer) that automatically learn distortion patterns from training data, replacing the traditional signal processing mechanism with an adaptive intelligent system
2Measurement precision
If machine learning models are trained on long audio signals, then distortion estimation accuracy is improved, but processing time increases for real-time applications
Solution Approach 1:
The patent applies segmentation by training the machine learning model on segmented portions of audio signals rather than entire long signals. The training data consists of multiple short segments, and the model learns to accurately estimate distortion from these smaller units, enabling real-time processing while maintaining accuracy through the cumulative learning from diverse segments
Solution Approach 2:
The patent performs preliminary action by pre-training the machine learning model offline on extensive training data before deployment. This preliminary training phase captures distortion patterns and relationships, so that during real-time operation, the already-trained model can quickly process new signals without requiring extensive computation, thus reducing processing time while maintaining high accuracy
Data Source
AI summary
A system includes a memory and a processor. The memory is configured to store a machine learning (ML) model. The processor is configured to (i) obtain a set of training audio signals in a form of a plurality of initial audio signals, which have first durations in a first range of durations and which are labeled with respective levels of distortion, (ii) train the ML model to estimate the levels of the distortion based on the training audio signals, (iii) receive an input audio signal having a duration in a second range of durations, shorter than the first durations, and (iv) estimate a level of the distortion in the input audio signal by applying the trained ML model to the input audio signal.