Neural Network Audio Sound Quality Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining sound quality of audio systems rely heavily on subjective human evaluation, requiring multiple expert assessors and lacking objectivity, while also being time-consuming and prone to variability.
Innovation Solution
A computer-implemented method using an artificial neural network that predicts sound quality scores by measuring frequency responses of audio systems, trained on datasets that include both frequency data and human evaluator scores, allowing for objective and efficient scoring without the need for multiple human evaluations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple human expert evaluators are used to determine sound quality, then objectivity is improved, but time consumption and cost increase
Solution Approach 1:
The patent creates a virtual copy of the human expert evaluation process through an artificial neural network model. The neural network is trained on datasets containing frequency response measurements and corresponding human evaluator scores, then uses this learned model to predict scores for new audio systems. This virtual copy replicates expert judgment without requiring actual human evaluators, resolving the contradiction between objectivity and time consumption.
Solution Approach 2:
The patent replaces the mechanical system of human evaluation (subjective listening and scoring by multiple experts) with an automated computational system. The neural network processes frequency response data objectively and consistently, eliminating the time-consuming human evaluation process while maintaining reliability through the objective nature of computational processing.
2Reliability
If multiple human expert evaluators are used to determine sound quality, then objectivity is improved, but variability and subjectivity remain
Solution Approach 1:
The neural network model creates a standardized virtual evaluator that applies the same evaluation criteria consistently to all audio systems. Unlike human evaluators who may have individual biases and varying interpretations, the neural network applies its learned evaluation function uniformly, eliminating variability while maintaining the objectivity of expert-level assessment.
3Measurement precision
If traditional frequency response measurement methods are used, then technical parameters are obtained, but they do not correlate well with subjective perception
Solution Approach 1:
The patent introduces an intermediary neural network model that bridges the gap between objective frequency response measurements and subjective perception. The neural network acts as a mediator, translating technical frequency domain data into predictions of human auditory perception by learning the complex relationship between physical measurements and subjective experience from training data.
Data Source
AI summary
A computer-implemented method for determining a scoring indicative of a sound quality of an audio system, the method comprising: sending at least one test signal to at least one reference audio system; measuring a first frequency response of the reference audio system to the test signal; receiving at least one reference scoring indicative of a sound quality of the reference audio system; supplying a training dataset comprising the first frequency response and the reference scoring to an artificial neural network; training the artificial neural network on the training dataset to predict a scoring for the audio system; sending at least one test signal to at least one production audio system; measuring a second frequency response of the production audio system; and processing an input dataset comprising the second frequency response by the artificial neural network to predict a scoring indicative of a sound quality of the production audio system.