Speech Emotion Recognition (SER) plays an important role in human-computer interface and assistant technologies. In this paper, a new method is proposed using distributed Convolution Neural Networks (CNN) to automatically learn affect-salient features from raw spectral information, and then applying Bidirectional Recurrent Neural Network (BRNN) to obtain the temporal information from the output of CNN. In the end, an Attention Mechanism is implemented on the output sequence of the BRNN to focus on target emotion-pertinent parts of an utterance. This attention mechanism not only improves the classification accuracy, but also provides model's interpretability. Experimental results show that this approach can gain 64.08% weighted accuracy and 56.41% unweighted accuracy for four-emotion classification in IEMOCAP dataset, which outperform previous results reported for this dataset.

In this study, we developed a novel heart rate (HR) monitoring approach in which we measure the pressure variance of the surface of the ear canal. A scissor-shaped apparatus equipped with a piezoelectric film sensor and a hardware circuit module was designed for high wearability and to obtain stable measurement. In the proposed device, the film sensor converts in-ear pulse waves (EPW) into electrical current, and the circuit module enhances the EPW and suppresses noise. A real-time algorithm embedded in the circuit module performs morphological conversions to make the EPW more distinct and knowledge-based rules are used to detect EPW peaks. In a clinical experiment conducted using a reference electrocardiogram (ECG) device, EPW and ECG were concurrently recorded from 58 healthy subjects. The EPW intervals between successive peaks and their corresponding ECG intervals were then compared to each other. Promising results were obtained from the samples, specifically, a sensitivity of 97.25%, positive predictive value of 97.17%, and mean absolute difference of 0.62. Thus, highly accurate HR was obtained from in-ear pressure variance. Consequently, we believe that our proposed approach could be used to monitor vital signs and also utilized in diverse applications in the near future.

Thermally and mechanically dependable two-dimensional (2D) photonic band gap lasers are proposed and realized at room temperature. The thin slab photonic band gap laser structure is sandwiched between air and a drilled aluminum oxide layer provided by wafer fusion techniques. In this thin slab structure, the optical confinement of photons is achieved by 2D triangular photonic lattice in horizontal plane and total internal reflection in vertical direction. Pulsed lasing action is observed at 1.54 μm by optical pumping with duty cycle up to 10%. Polarization characteristics of two-dimensional photonic band gap defect modes are also studied below and above the lasing threshold.

No abstract

We fabricated a photonic band edge laser array based on a two-dimensional square-lattice photonic crystal (PC) slab waveguide using a laser holography (LH) method, instead of the commonly used electron-beam lithography (EBL). The nature of the LH process enabled high-throughput large-area fabrication of band edge lasers. Moreover, the laser performance was comparable to that of reported EBL counterparts. Careful examination of the spectral positions of the observed modes with respect to the calculated photonic band structure identified three main band edge modes as the origins of lasing: M2, X2, and M1. Owing to the gradual change in the air-hole size of the PC, the lasing modes shifted monotonically along the laser array, resulting in an M1 mode span of ∼30nm (centered at 1550nm) over a distance of 5mm.