In the past decade, improving upper limb prostheses control methods with pattern recognition (PR) has been the focus of an extended amount of research. However, several challenges associated with the processing of the Electromyogram (EMG) signals still need to be tackled to enable widespread and clinical implementation of upper limb prostheses with PR. As a result, alternative modalities functioning as promising control signals have been proposed as source of control input rather than the surface EMG, such as Acoustic myography (AMG) and Force myography. In this paper, eight high sensitivity array microphones were utilized to acquire the AMG signals, with 8 custom-built 3D printed microphone housing developed for the purpose of this research. Twenty subjects were recruited for data collection in this paper with the hardware design developed specifically by our research team, making our database the largest open-access dataset in the AMG literature. We proposed a novel feature extraction (FE) method based on the Wavelet Scattering Transform (WST) to tackle the challenge of extracting the relevant information from AMG to classify 14 hand and finger movement classes. The WST is a translation-invariant non-linear signal representation that has a strong theoretical support as it maintains stability to time-warping deformations, while preserving a high degree of discriminability. The performance results showed that WST outperformed all state-of-the-art FE methods with an accuracy of 88% on average across 20 subjects when classified with Quadratic Discriminant Analysis (QDA) classifier for a large dataset of AMG signals. These results suggest that the AMG signals can be utilized as a reliable source of control, especially when the windows sizes and number of channels are carefully selected. The AMG dataset is available from the link https://drive.google.com/drive/folders/1r0rBnrNG5c8qegffKUhAYYHQUT8cGVHD?usp=sharing .

In environment sound classification, log Mel band energies (MBEs) are considered as the most successful and commonly used features for classification. The underlying algorithm, fast Fourier transform (FFT), is valid under certain restrictions. In this study, we address these limitations of Fourier transform and propose a new method to extract log Mel band energies using amplitude modulation and frequency modulation. We present a comparative study between traditionally used log Mel band energy features extracted by Fourier transform and log Mel band energy features extracted by our new approach. This approach is based on extracting log Mel band energies from estimation of instantaneous frequency (IF) and instantaneous amplitude (IA), which are used to construct a spectrogram. The estimation of IA and IF is made by associating empirical mode decomposition (EMD) with the Teager–Kaiser energy operator (TKEO) and the discrete energy separation algorithm. Later, Mel filter bank is applied to the estimated spectrogram to generate EMD-TKEO-based MBEs, or simply, EMD-MBEs. In addition, we employ the EMD method to remove signal trends from the original signal and generate another type of MBE, called S-MBEs, using FFT and a Mel filter bank. Four different datasets were utilised and convolutional neural networks (CNN) were trained using features extracted from Fourier transform-based MBEs (FFT-MBEs), EMD-MBEs, and S-MBEs. In addition, CNNs were trained with an aggregation of all three feature extraction techniques and a combination of FFT-MBEs and EMD-MBEs. Individually, FFT-MBEs achieved higher accuracy compared to EMD-MBEs and S-MBEs. In general, the system trained with the combination of all three features performed slightly better compared to the system trained with the three features separately.

Acoustic myography (AMG) measures the pressure waves produced by muscle contraction withmicrophones coupled to small air cavities placed on the skin. This paper investigates whetherforearm AMG signals constitute a biometric trait, using the methodology of speakerrecognition. The AMG signal is modelled as the response of a subject-dependent biomechanicalchannel to a neuromuscular excitation, a source--channel model that guides the analysis of theorigin of the identity information. The study relies on three public datasets recorded from45 participants with three systems. Cepstral features derived from the model and deepembeddings are evaluated under the same protocols: gesture-dependent and gesture-independentrecognition, recordings without voluntary movement, inter-session evaluation, subjects unseenduring training and open-set identification. In the gesture-dependent protocol, rank-1identification rates of 99.3--99.8\% and equal error rates (EER) of 1.7--4.4\% are obtained withthe cepstral features, and below 1\% with the deep embeddings. Identity is also recognised fromrecordings without voluntary movement (EER 1.1--4.2\%), and persists across days when the braceletis placed again (EER 7.2\%). It transfers to unseen gestures with degraded performance(13.5--22.6\%), but not between gestures and rest. For subjects unseen during training, the EER is3.5--6.7\% with both representations on the two larger datasets. Identity information is mainly carriedby the shape of the spectral envelope rather than by the signal level or the temporal dynamics.These findings support the hypothesis of a subject-specific biomechanical signature whoserepresentation depends on the excitation, and establish forearm AMG as a candidate modality forwearable biometrics.

To improve quality of life for individuals with hand amputation, active upper-limb prostheses aim to restore hand function through reliable, intuitive control. Electromyography remains the dominant sensing modality for gesture-based prosthetic control, but its robustness is often limited by signal variability and user-specific calibration needs. This paper investigates acoustomyography (AMG) as an alternative sensing modality for real-time prosthetic hand control, motivated by the mechanical nature of AMG signals, which reflect muscle-generated acoustic activity and may offer more stable sensing across users. Statistical analyses show AMG signals remain stationary during gesture execution, supporting window-based real-time processing, while measured inter-subject dependency remains below 5%, highlighting AMG's potential for user-independent, calibration-free interfaces. To validate this, a complete embedded real-time control system was developed, integrating a forearm-worn multi-sensor bracelet for AMG acquisition, a 5-DOF prosthetic hand, and an STM32 microcontroller for TinyML inference and hand control. Analysis of acquisition window duration evaluated the trade-off between accuracy and control latency: longer windows improve recognition but increase response delay. Windows between 200 and 400 ms with 50% overlap give the best real-time compromise. The system achieves 92.1% accuracy with 245 ms total delay using a 300 ms window, and 95.0% accuracy with 318 ms delay using a 400 ms window. It further achieves 95.0% accuracy under subject-mixed evaluation and 90.6% under leave-one-subject-out cross-validation, a subject-dependency index of only 4.63%. Crucially, real-time evaluation on five subjects unseen during training confirms generalization, yielding 95.3% (±1.7%) accuracy with no subject-specific adaptation.