Automatic Guitar Transcription With Deep Neural Networks
Guitar tablature is a form of musical notation designed specifically for fretted string instruments. It is crucial for musicians, especially beginners, as it provides an easy-to-read format that visually represents the placement of the fingers on the instrument, making it more accessible to learn an...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11059978/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Guitar tablature is a form of musical notation designed specifically for fretted string instruments. It is crucial for musicians, especially beginners, as it provides an easy-to-read format that visually represents the placement of the fingers on the instrument, making it more accessible to learn and play songs accurately. Automatic transcription of guitar music into tablature is a complex task in Music Information Retrieval, which has been enhanced by recent advances in deep learning. This article builds on a state-of-the-art note-level transcription model that uses a self-attention mechanism. The model incorporates beat-informed quantisation to accurately convert audio signals into tablature, overcoming challenges such as determining where notes are played on the guitar neck. In replicating this model, its properties have been investigated, and the results have been analysed. A key aspect of this article is the development of a new dataset designed to assess the robustness of the model in different scenarios. This is crucial due to the limited availability of data in this area. Through extensive experimentation and analysis, this study evaluates the performance of the model on unseen data, identifying its strengths and areas for improvement. In addition, this article provides insights into the mechanism of self-attention and its effectiveness in tasks such as Automatic Music Transcription. The model was tested with multiple attention heads to study their impact on performance, but this modification did not show significant improvement. Therefore, it suggests that other areas, such as improving the quality and quantity of data, may be more crucial to improve performance. |
|---|---|
| ISSN: | 2169-3536 |