Abstract
Automatically detecting the playing styles of musical instruments could assist in the development of intelligent software for music coaching and training. However, the respective methodologies are still at an early stage, and there are limitations in the playing techniques that can be identified. This is partly due to the limited availability of complete and real-world datasets of instrument playing styles that are mandatory to develop and train robust machine learning models. To address this issue, in this data article, we introduce a multimodal dataset consisting of 549 video samples in MP4 format, and their respective audio samples in WAV format, covering nine different electric guitar techniques in total. These samples are produced by a recruited guitar player using a smartphone device. The recording setup is designed to closely resemble real-world situations, making the dataset valuable for developing intelligent software applications that can assess the playstyle of guitar players. Furthermore, to capture the diversities that may occur in a real scenario, different exercises are performed using each technique with three different electric guitars and three different simulation amplifiers using an amplifier simulation profiler. In addition to the audio and video samples, we also provide the musescores of the exercises, making the dataset extendable to more guitar players in the future. Finally, to demonstrate the effectiveness of our dataset in developing robust machine learning models, we design a Support Vector Machine (SVM) and a Convolutional Neural Network (CNN) for classifying the guitar techniques using the audio files of the dataset. The code for the experiments is publicly available in the dataset's repository.