ML Audio Quality Assessment With Human-Like Degradation Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The use of human assessors for determining mean opinion scores in audio quality assessment is costly and time-consuming, making it impractical for prompt and cost-effective evaluations.
Innovation Solution
A machine learning model, such as a neural network, is trained to estimate human audio quality assessments by comparing ML-generated scores to human assessments, using feedback to refine its performance, and is used to improve an audio enhancement model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human assessors are used to determine mean opinion scores, then measurement precision of audio quality is improved, but loss of time and cost increase
Solution Approach 1:
The patent creates a machine learning model that copies human assessors' evaluation capabilities. The system trains neural networks to replicate human subjective quality assessments by feeding them training data consisting of audio samples paired with human mean opinion scores. Once trained, the ML model can generate quality assessments that closely mimic human judgments without requiring actual human listeners, thus resolving the contradiction between measurement precision and time loss.
Solution Approach 2:
The patent replaces the mechanical system of human assessment (requiring human time and effort) with an automated electronic system. The machine learning model processes audio signals through computational algorithms to generate quality scores, substituting the biological human assessment mechanism with an electronic computation mechanism that operates faster and at lower cost while maintaining assessment accuracy.
2Measurement precision
If human assessors are used to determine mean opinion scores, then measurement precision of audio quality is improved, but cost increases
Solution Approach 1:
The patent creates a machine learning model that copies human assessors' evaluation capabilities. The system trains neural networks to replicate human subjective quality assessments by feeding them training data consisting of audio samples paired with human mean opinion scores. Once trained, the ML model can generate quality assessments that closely mimic human judgments without requiring actual human listeners, thus resolving the contradiction between measurement precision and time loss.
Solution Approach 2:
The patent employs machine learning models that are computationally inexpensive compared to human assessors. The system uses trained neural networks that can be deployed on standard computing infrastructure, replacing expensive human resources with affordable computational resources. The ML model provides a cost-effective alternative that maintains assessment precision while dramatically reducing the cost per evaluation.
3Productivity
If machine learning model is used for audio quality assessment, then productivity is improved, but measurement precision may worsen
Solution Approach 1:
The patent applies preliminary action by training the machine learning model extensively before deployment. The system uses a training phase where the ML model learns from numerous examples of audio samples paired with human quality assessments. This preliminary training equips the model with the knowledge to make accurate predictions, ensuring that when the model is used for actual assessments, it maintains high measurement precision while achieving high productivity.
Solution Approach 2:
The patent incorporates feedback mechanisms where the machine learning model's predictions are compared against actual human assessments during training and validation. The system uses this feedback to iteratively improve the model's accuracy through techniques like loss minimization and parameter optimization. This feedback loop ensures that the model achieves both high productivity and maintained measurement precision.
Data Source
AI summary
An audio assessment machine learning model, for example including one or more neural networks, may be trained for human audio quality assessment estimation, wherein the training comprises performing comparisons of a plurality of training machine learning audio quality assessments of training audio content to a plurality of human audio quality assessments of the training audio content and adjusting the audio assessment machine learning model based on the comparisons. After the training, a first audio analysis of first audio content may be performed by the audio assessment machine learning model. A first machine learning audio quality assessment of the first audio content may be provided, by the audio assessment machine learning model, based on the first audio analysis. The first machine learning audio quality assessment may include a quality score for the first audio content and a quality degradation reason for the first audio content.