Implementation and Analysis of Algorithms for Pitch Estimation in Musical Fragments

The paper presents an analysis of existing algorithms for single-pitch estimation and proposes a new neural network based algorithm for musical chords recognition. To analyze the algorithms their software implementation was performed and used for the recognition of notes in melodies containing noise echoes and overtones, as well as specific playing techniques characteristic of certain musical instruments. Based on the obtained results a comparative analysis of the existing algorithms is performed specifying their limitations and disadvantages which impede their efficient application for chord recognition. The implemented algorithm based on a fully connected three-layer neural network for chord recognition is described. Assessment of its accuracy and efficiency is shown.