Skip to Main Content
This paper presents a method to transcribe polyphonic music based on adaptive comb filter and neural network. In the method, the input audio is firstly divided into snapshots by a BP neural network, and then comb filters of different notes are used to calculate features. Comb filter can be adjusted to take inharmonicity and note pitch error into consideration. Note energy, audible harmonic number and harmonic continuity obtained from filter are used in BP neural network to recognize notes in snapshot, which helps multipitch estimation algorithm to be more robust to frequency missing and sharing.