Although audio-augmented reality (AAR) has known applications in music, the use of wearables such as augmented reality (AR) glasses for egocentric audio data capture for music has not been investigated. Current egocentric datasets are mostly focused on speech research, neglecting music's unique demands for tasks such as real-time optimisation or assistive listening. This paper introduces EgoMusic, a multimodal dataset featuring synchronised egocentric audio-visual data captured with AR glasses during live performances, alongside studio-quality audio references. We investigate AR glasses' utility for music and baseline artificial intelligence (AI) approaches for hearing enhancement, positioning EgoMusic as the first dataset that enables research for egocentric music AAR.

Audio-based health monitoring systems have gained traction for its capability in capturing different physiological target signals and holding features relevant in diagnosis and treatment, making it a feasible form of data acquisition in biomedical signal processing and feature extraction. Acquiring audio for both heart and respiratory sounds for rate estimation was accomplished using the bottom microphone of the Google Pixel 6A from the trachea and anterior left. A direct comparison of the filter-Hilbert and wavelet-Shannon algorithms based on literature with modifications catering to the application of this study was done to measure their performance and reliability against the KardiaMobile 6L EKG. Tests done were able to show good accuracy for both algorithms on heart rate calculations on both the trachea and anterior left, satisfying having less than ± 10 BPM difference from control. Tests for respiratory rate only provided promising results of less than ± 5 CPM difference from control for tracheal recordings,

This study contributes to improving heart rate (HR) monitoring by comparing the Peripheral Beat Amplitude (PBA) and Smoothed Z-score Peak Detection (Z-Score) algorithms for HR extraction across different body sites. The Z-Score algorithm, leveraging on edge computing, demonstrated higher accuracy than PBA, with a Mean Absolute Error (MAE) around 4 BPM. The study also explores the suitability of the Perfusion Ratio technique for Blood Oxygen (SpO2) levels and the Welch method for respiratory rate (RR). Testing was conducted simultaneously at four distinct body sites (finger, wrist, arm, and ankle) during sitting and walking activities on six healthy individuals. Results indicate that the wrist is the optimal measurement site for various physiological parameters. Post-processing techniques were deliberately chosen to enhance the system's overall performance. The focus on HR necessitated evaluating two algorithms, given its critical importance. For SpO2and RR, single algorithms were selected to streamline processing while ensuring reliable measurements. The developed system exhibits acceptable power consumption of 51.98 m W and achieves an average coverage of 80% in controlled tests, indicating a significant portion of monitoring time provides accurate physiological data. This study emphasizes the deliberate selection of algorithms and measurement sites, demonstrating the superiority of the Z-Score algorithm and the practicality of integrating HR, SpO2, RR, and PPG monitoring in a single system.

Music transcription plays a significant role in the music community in terms of learning and sharing knowledge about musical pieces. However, for guitar tablatures, most existing transcription systems fail to incorporate articulation detection. In this study, an automatic guitar transcription (AGT) system, which uses a monophonic guitar recording as input to detect and identify the string-fret combinations and articulations (legato) played, was developed. Algorithms for each system block were chosen and modified to fit the system specifications. Results show that the modifications led to improvements in the string-fret block accuracy, from 78% to 87%, and the articulation block F-measure, from 59% to 84%. The AGT system was also compared with a commercial music transcription application. While both were trained on different data sets, the AGT system outperformed the latter, with the system having 78.65% string-fret accuracy and 93.23% articulation accuracy compared to the commercial application's 48.44% string-fret accuracy and 70.31% articulation accuracy.

The plucking point position is an important gestural parameter that controls the tone produced in a classical guitar. Existing studies used a theoretical model of the plucking point position to estimate the parameter from audio signals. However, in performances where the classical guitarist evokes different tonal characteristics through varying gestural parameters, a theoretical model may not suffice. In this work, we explored the use of a novel video-based approach in estimating the parameter and compared it to the more common audio-based approach. We collected a data set consisting of real classical guitar performances from a classical guitarist performing with his plucking nails as the medium of excitation. We have discovered that on our data set, the video-based approach, which had an average absolute error of 1.0656 cm, outperformed the audio-based approach, which had an average absolute error of 2.5904 cm, in terms of the estimation accuracy. In terms of the speed however, the audio-based approach had a faster estimation time (13.63 milliseconds) on average compared to the video-based approach (176.96 milliseconds). Hence, we recommend the video-based approach for applications where the estimation accuracy is critical, especially on performances with varying tonal characteristics due to different gestural parameters. Meanwhile, we recommend the audio-based approach in real-time applications.

The classical guitar tone can be controlled by a number of gestural parameters performed by the player, including the plucking point position, the angle of attack, and the angle of release. In this study, we developed a video-based approach for estimating the angle of attack from a classical guitar performance. Our angle of attack estimation system achieved an average absolute error of 3.32 degrees on our data set containing 270 samples with varying angles of attack and plucking point positions. Moreover, we discovered through statistical analysis that the two gestural parameters indeed significantly affect the audio feature that describes the degree of brightness of the produced guitar sound, namely the spectral centroid. We were able to generate regression models with fair regression scores that correlate the spectral centroid with the two gestural parameters, and we further discovered that the regression scores decrease with increasing fundamental frequency of the classical guitar sound.