A Hybrid Recurrent Neural Network For Music Transcription

TL;DR

This paper introduces a hybrid RNN-based model that integrates music language models with acoustic classifiers to enhance automatic music transcription accuracy, demonstrating superior performance on piano datasets.

Contribution

The paper presents a novel generative architecture combining RNN-based music language models with acoustic classifiers for improved transcription accuracy.

Findings

The proposed model outperforms existing methods on the MAPS piano dataset.

Incorporating higher-level score information improves transcription performance.

Different neural network architectures for acoustic modeling were compared.

Abstract

We investigate the problem of incorporating higher-level symbolic score-like information into Automatic Music Transcription (AMT) systems to improve their performance. We use recurrent neural networks (RNNs) and their variants as music language models (MLMs) and present a generative architecture for combining these models with predictions from a frame level acoustic classifier. We also compare different neural network architectures for acoustic modeling. The proposed model computes a distribution over possible output sequences given the acoustic input signal and we present an algorithm for performing a global search for good candidate transcriptions. The performance of the proposed model is evaluated on piano music from the MAPS dataset and we observe that the proposed model consistently outperforms existing transcription methods.