Advancing Audio Fingerprinting Accuracy in Challenging Environments: a Hybrid Approach Combining Traditional Methods and AI Techniques

EasyChair Preprint 14669

Abstract

Audio fingerprinting serves as a fundamental technology for identifying and matching audio content across various platforms, from music recognition to copyright enforcement. However, maintaining accuracy in challenging environments—characterized by high levels of noise, compression artifacts, or signal distortions—remains a significant challenge. This abstract proposes a hybrid approach that combines the strengths of traditional audio fingerprinting methods with advanced artificial intelligence (AI) techniques to enhance accuracy and reliability under adverse conditions.

The research begins by analyzing the limitations of traditional audio fingerprinting methods, such as the reliance on specific features like spectral peaks, which can be susceptible to environmental noise and audio compression. These methods, while computationally efficient, often fail to maintain accuracy when the audio signal undergoes significant modifications. To address these limitations, the study explores the integration of AI techniques, particularly deep learning models, to complement and reinforce traditional fingerprinting approaches.

A key innovation in this research is the development of a hybrid framework that utilizes conventional feature extraction techniques alongside AI-driven models. The traditional methods are employed to extract robust initial features from the audio signal, such as spectral peaks, which are then fed into a deep learning model. This model, typically a convolutional neural network (CNN) or a recurrent neural network (RNN), is trained to enhance these features and generate more resilient fingerprints that are less susceptible to noise and distortions.

Keyphrases: Artificial Intelligence, Convolutional Neural Networks (CNNs), Hybrid Approach, Traditional methods, Transfer Learning, audio fingerprinting, audio signal processing, deep learning, distortion resistance, noise robustness, real-time audio identification

@booklet{EasyChair:14669,
  author    = {Oluwaseyi Oladele},
  title     = {Advancing Audio Fingerprinting Accuracy in Challenging Environments: a Hybrid Approach Combining Traditional Methods and AI Techniques},
  howpublished = {EasyChair Preprint 14669},
  year      = {EasyChair, 2024}}