A Bi-directional Transformer for Musical Chord Recognition

Abstract

Chord recognition is an important task since chords are highly abstract and descriptive features of music. For effective chord recognition, it is essential to utilize relevant context in audio sequence. While various machine learning models such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs) have been employed for the task, most of them have limitations in capturing long-term dependency or require training of an additional model.In this work, we utilize a self-attention mechanism for chord recognition to focus on certain regions of chords. Training of the proposed bi-directional Transformer for chord recognition (BTC) consists of a single phase while showing competitive performance. Through an attention map analysis, we have visualized how attention was performed. It turns out that the model was able to divide segments of chords by utilizing adaptive receptive field of the attention mechanism. Furthermore, it was observed that the model was able to effectively capture long-term dependencies, making use of essential information regardless of distance.

Improving the Classification of Rare Chords With Unlabeled Data

“…We ran our experiments using the previously stated state-ofthe-art classifier [15] and implemented the necessary modifications in order to run both the focal loss and self-learning examples. For our labeled dataset, we used the Isophonics Queen and Beatles dataset [2], and as our (large) unlabeled data, we used the audios indicated by the DALI dataset [18] which results in around 5,000 songs without chord label annotations.…”

“…Major and minor chords still dominate, but that is inevitable since they tend to accompany the rare chords. It is also important to notice that the generated subset has a smaller variety of chords, as the classifier we used, a state-of-the-art ACR technique based on the Transformer architecture [15], has a limited number of classes it is able to predict. If selectedDuration ≥ desiredDuration then move to next chord type Another important component of [13] is the addition of noise to the selected subset.…”

Jointist: Joint Learning for Multi-instrument Transcription and Its Applications

“…Transformer networks [37] have been shown to work well for a wide range of MIR tasks [38][39][40][41][42][43][44]. In this paper, we adopt the music tagging transformer proposed in [44] as our musical instrument recognition module, f IR .…”

nnAudio: An on-the-fly GPU Audio to Spectrogram Conversion Toolbox Using 1D Convolution Neural Networks

“…WaveNet is a model designed to take raw waveforms as input, and has inspired several recent audio related machine learning models [4][5][6]. Despite these advances, countless models are still using frequency domain features as the model's input for various tasks due to their superior performance [7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25]. Therefore, there is still value in developing a faster timefrequency conversion computation method, which is what we propose in this paper.…”