Construction of piano audio fingerprint based on local maximum chromatic energy combined with note onset

Abstract

With the rapid growth of audio information resources, the efficient management and accurate retrieval of massive music data have become the focus of research. To improve the accuracy and efficiency of music data retrieval, a method combining local maximum chromaticity energy with note onset detection is proposed. The piano audio fingerprint is constructed by extracting key features and generating digital identifiers. The innovation of this method lies in using note start detection to capture transient features in audio signals and combining local maximum chromaticity energy points to enhance the uniqueness and robustness of audio fingerprints. The experimental results showed that the overall recognition accuracy of the constructed fingerprint technology performed well in different dataset sizes, remaining around 93%. The recognition accuracy fluctuated around 93% in general and decreased slightly when the dataset size was larger than 400. When the dataset was 1,000, the recognition accuracy was about 90%. In terms of the audio fingerprint extraction effect, the technology had a maximum fingerprint distance of 0.80 for the songs of For Forever and Shape of You, and a minimum fingerprint distance of 0.56 for Despacito and For Forever. In terms of the performance of the fingerprint retrieval system, the research system had a minimum hit rate of 84 and a maximum hit rate of 112, while the retrieval accuracy was mostly above 90%. The audio fingerprint method was superior in recognition accuracy, robustness, and scalability. The study provides technical support and a research basis for the application of audio watermarking, copyright protection, and audio classification.