Audio Distortion Quantification Using Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio systems lack an effective method to quantify and measure distortion, particularly for short-duration audio signals, which limits their ability to design more linear systems and control intentional distortion in real-time.
Innovation Solution
A system utilizing a machine learning model, such as a convolutional neural network (CNN) or generative adversarial network (GAN), to convert audio signals into images and estimate the level of distortion, including Total Harmonic Distortion (THD), even for very short audio samples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional distortion sensing methods are used, then distortion can be detected in audio signals, but the method cannot effectively quantify distortion in short-duration audio signals
Solution Approach 1:
The patent transforms one-dimensional audio signals into two-dimensional images by converting time-domain waveforms into spatial representations. This dimensional transformation enables the use of convolutional neural networks, which are specifically designed to process image data, thereby achieving accurate distortion quantification in short-duration audio signals that were previously unmeasurable with traditional methods
2Measurement precision
If machine learning models are trained on long audio signals, then the model can accurately estimate distortion, but the model cannot process short-duration audio signals
Solution Approach 1:
The patent segments long audio signals into shorter clips during the training process, creating a dataset where each training sample is a short-duration audio segment. This segmentation allows the machine learning model to learn distortion patterns from short signals during training, enabling it to accurately process and quantify distortion in short-duration audio signals without requiring the entire audio file to be long
3Productivity
If traditional audio analysis methods are used, then the system structure remains simple, but the system cannot provide real-time control of distortion
Solution Approach 1:
The patent replaces traditional mechanical/audio-based distortion measurement systems with a machine learning-based system. By using pre-trained convolutional neural networks that can process audio images in real-time, the system achieves fast distortion quantification and control without requiring complex real-time signal processing hardware, thereby enabling real-time control while maintaining reasonable system complexity
Data Source
Figure 1
Figure 2
Figure 3
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 that are labeled with respective levels of distortion, (ii) convert the training audio signals into respective images, (iii) train the ML model to estimate the levels of the distortion based on the images, (iv) receive an input audio signal, (v) convert the input audio signal into an image, and (vi) estimate a level of the distortion in the input audio signal, by applying the trained ML model to the image.