Violin Note Spectrum Detection Based on a Multi-Fundamental Frequency Estimation Algorithm

Abstract

Traditional fundamental frequency estimation methods often exhibit misjudgments when processing polyphonic violin music due to harmonic overlaps, which significantly limit the accuracy of automatic music transcription systems. To address this challenge, this paper proposes a method for detecting violin note spectra based on a multi-fundamental frequency estimation algorithm. First, to mitigate the interference of fundamental information in the cepstrum, a note-corrected reverse-banding process is introduced. This approach enhances cepstral peaks while suppressing high-frequency noise, thereby improving the accuracy of fundamental period recognition. Second, a multi-resolution rapid time– frequency analysis method (RTFI) is employed for harmonic extraction, effectively separating overlapping harmonic components and improving the precision of fundamental frequency estimation. Finally, given the temporal variability of note spectral features, a single-frame multi-fundamental-frequency phased-estimation method is developed. This method separately estimates the transient and steady-state stages of each note, further enhancing the accuracy of multi-fundamental frequency estimation. Experimental results demonstrate that, across tests involving one to nine notes, the proposed algorithm outperforms existing approaches such as HPS, ISSA, and JEA in terms of recall, precision, and F-measure metrics. Notably, under the complex scenario of six simultaneous notes, the proposed algorithm achieves an F-measure of 94%, significantly exceeding those of the comparison methods. In addition, the proposed method shows superior performance in fundamental frequency count estimation and note label recognition, with a markedly lower total error rate.