International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 12 | Dec 2025
p-ISSN: 2395-0072
www.irjet.net
Advancing Audio Fingerprinting accuracy with AI & ML Prof. Shobha S Biradar1, Soumya Nasi2 1
Professor, Master of Computer Application, VTU, Kalaburagi , Karnataka ,India Student, Master of Computer Application, VTU, Kalaburagi , Karnataka ,India ------------------------------------------------------------------------------***------------------------------------------------------------------Recent advances in Artificial Intelligence (AI) and Machine ABSTRACT- Audio fingerprinting has become a 2
Learning (ML) are transforming audio fingerprinting into a more robust and intelligent technology. Deep learning models, particularly Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Transformers, have demonstrated remarkable ability to extract high-level, noiseinvariant features from audio signals. Moreover, selfsupervised learning (SSL) frameworks such as wav2vec2.0 and Hubert enable the learning of rich representations without requiring large labeled datasets. Metric learning techniques, including Siamese and Triplet networks, further enhance accuracy by mapping similar audio signals closer together in the embedding space, regardless of distortions.
fundamental technology for music recognition, copyright protection, and audio forensics. Traditional approaches rely on handcrafted signal processing techniques such as spectrogram analysis, Chroma features, and MFCCs, which often suffer from reduced accuracy under noisy conditions, pitch or tempo variations, and overlapping audio sources. Recent advances in Artificial Intelligence (AI) and Machine Learning (ML) have significantly improved the robustness and precision of audio fingerprinting systems. Deep learning models, including Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Transformers, enable extraction of noise-invariant and semantically rich features from audio signals. Self-supervised learning frameworks further enhance representation quality without the need for extensive labeled data. Metric learning techniques, such as Siamese and Triplet networks, provide powerful embedding spaces for efficient and accurate audio matching. Hybrid architectures that integrate traditional signal processing with deep embeddings also address scalability challenges in large-scale music databases.
The integration of these AI-driven methods addresses key limitations of traditional systems, offering greater resilience, scalability, and adaptability. As a result, audio fingerprinting is evolving into a versatile and intelligent solution capable of supporting applications at global scale. However, challenges remain in balancing accuracy, computational efficiency, and real time performance for deployment in large-scale and resource-constrained environments.
Keyword: Audio Fingerprinting, Artificial Intelligence, Machine Learning, Deep Learning, Spectrogram, Feature Extraction, Metric Learning, Audio Recognition, SelfSupervised Learning, Audio Forensics.
2. PROBLEM STATEMENT Noise Sensitivity: Background noise, reverberation, and environmental distortions significantly reduce recognition accuracy. Transformation Vulnerability Pitch shifts, tempo changes, or compression techniques can alter the fingerprint, leading to mismatches. Overlapping Audio Sources: When multiple sounds occur simultaneously, traditional methods struggle to extract reliable fingerprints. Scalability Issues: Searching for matches in massive music or audio databases is computationally challenging, often resulting in slower retrieval times. Lack of Robust Representation: Handcrafted features fail to capture high-level semantic characteristics of audio, limiting robustness and adaptability.
1. INTRODUCTION Audio fingerprinting is a powerful technique that enables the identification of audio content by generating compact, distinctive digital representations of sound signals. Unlike raw audio files, fingerprints are lightweight, resilient to distortions, and uniquely associated with specific recordings. This makes fingerprinting essential in diverse applications such as music recognition, copyright enforcement, broadcast monitoring, audio forensics, and smart devices.
3. OBJECTIVES
Conventional fingerprinting systems largely rely on signal processing techniques such as spectrogram peak detection, Chroma features, and Mel-frequency costrel coefficients (MFCCs). While these approaches are efficient and computationally lightweight, they face significant challenges when audio undergoes transformations like background noise, pitch and tempo alterations, reverberation, or overlapping sources. Such limitations reduce their accuracy and reliability in real-world conditions where audio rarely exists in pristine form.
© 2025, IRJET
|
Impact Factor value: 8.315
The primary objective of this study is to design and implement an AI- and ML-enhanced audio fingerprinting system that achieves higher accuracy, robustness, and scalability compared to traditional approaches To develop deep learning-based feature extraction models (e.g., CNNs, RNNs, Transformers) capable of generating noise-invariant and transformation-resilient audio fingerprints. To apply metric learning techniques (e.g., Siamese and Triplet networks) for creating embedding spaces that enable
|
ISO 9001:2008 Certified Journal
|
Page 536