ResNet-based audio quality evaluation model and application

Through the audio quality evaluation model based on ResNet, audio data is processed using Fourier transform and Mel frequency scale, and features are extracted through ResNet, and the audio quality score results are finally output through the classifier, which solves the problem of insufficient accuracy and efficiency in complex audio scenarios, achieving more efficient and accurate audio quality evaluation.

CN119763601APending Publication Date: 2025-04-04UNIV OF SHANGHAI FOR SCI & TECH
0 Cites 0 Cited by

Patent Information

Application Number
CN202411897083.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-12-23
Publication Date
2025-04-04

AI Technical Summary

Technical Problem

When traditional audio quality evaluation methods deal with complex and changeable audio scenarios, it is difficult to meet the needs of high precision and high efficiency, especially in the quality evaluation of singing audio.

Method used

The audio quality evaluation model based on ResNet is used, and the frequency information after Fourier transform is mapped to the Mel frequency scale, and then converted into a logarithmic amplitude spectrum to obtain a spectrum map close to the perception method of the human ear. The spectrum graph is divided into local feature areas, sent to the ResNet module to extract features, and the audio quality score results are output through the classifier.

Benefits of technology

By combining residual connections and two-dimensional convolutional convolution layers, the network can deeply mine the texture features of the audio, keeping the details of the features not lost as the network goes deeper, thereby improving the efficiency and accuracy of audio quality evaluation.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention relates to the technical field of audio quality evaluation models, and particularly discloses an audio quality evaluation model based on ResNet and application, and the evaluation model is constructed according to the following steps: S1, mapping frequency information of audio data after Fourier transform to a Mel frequency scale, converting the Mel frequency scale into a logarithmic magnitude spectrum, obtaining a spectrogram close to a human ear perception mode; s2, dividing the spectrogram into a series of local feature regions with the size of W * H, and reshaping each region into a feature vector X with the size of C * W * H; according to the method, the convolutional neural network with a residual structure is adopted, and modeling is carried out for MFCC audio features. Through combination of residual connection and a convolutional layer of two-dimensional convolution, the network can deeply mine textural features of the audio. Meanwhile, the output of the previous layer and the output of the next layer are connected and spliced through the residual error, and the details of the features are not lost along with the deep network. Therefore, the audio quality evaluation efficiency and accuracy are improved.
Need to check novelty before this filing date? Find Prior Art