Search

The end-to-end speech recognition technology solves the problem that each component is independent and models cannot be jointly optimized in the traditional speech recognition model. It incorporates such components as the acoustic model, language model, and decoding unit of the hybrid model into a single neural network, that can avoid the inherent defects of multiple modules and greatly reduces the complexity of the speech recognition model. In this research, an Amdo-Tibetan speech recognition system is constructed based on Listen, Attend and Spell (LAS) model by the end-to-end speech recognition technology. It can realize the direct conversion from Amdo-Tibetan speech sequence to the corresponding character sequence and greatly reduces the difficulty of building the Amdo-Tibetan speech recognition model. To further improve the performance of the proposed system, the following improvements have been made: firstly, the Multi-Head Attention mechanism is introduced to improve the alignment accuracy between state vectors of decoder and encoder; secondly, the label smoothing technique is adopted to solve the problem of over-fitting; thirdly, an N-gram language model is combined with the LAS model to increase the accuracy of speech recognition and the maximum mutual information (MMI) criterion is employed for discriminative training; and finally, transfer learning is utilized to overcome the problem of insufficient training data. Experimental results show that the proposed model can significantly enhance the performance of Amdo-Tibetan speech recognition.

This paper investigates the intensity noise and pulse oscillation characteristics of an InAs/GaAs quantum dot laser epitaxially grown on germanium. We show that the relative intensity noise of the free-running laser generally decreases with increasing pump current, and the minimum value reaches down to about $-$ 126 dB/Hz. The intensity noise is hardly affected by the optical feedback, unless there is a resonance or pulse oscillation in the noise spectrum. The laser pumped at a high current is more sensitive to the optical feedback. Interestingly, it is found that the free-running Ge-based quantum dot laser generates self-sustained pulse oscillations with one period or two periods upon the pump current, without incorporating saturable absorbers. This behavior is valuable for both self-generation of photonic microwaves and for understanding nonlinear dynamics of semiconductor lasers.

Condition monitoring plays an important role in wind turbine maintenance, wherein monitoring turbines’ unhealthy statuses, mainly those not-runnable statuses causing turbines stop working, could be beneficial in both maintenance cost and wind power generation. A data-driven approach based on support vector data description (SVDD) and extreme learning machine (ELM) algorithms is proposed in this paper to realize effective monitoring on wind turbines’ unhealthy status. The SVDD algorithm is applied to separate data of unhealthy statuses from the healthy one. The ELM algorithm is used to construct a classifier for monitoring unhealthy (not-runnable) statuses of wind turbines. Industrial data from real wind farms are studied. Numerical results illustrate the feasibility of the proposed approach, and comparison studies with six monitoring algorithms validate that the proposed two-phase model could obtain better performance than other models.

This letter reports on a 1.3- $\mu \text{m}$ reflection insensitive transmission with a quantum dot laser directly grown on silicon in the presence of strong optical feedback. These results show a penalty-free transmission at 10 GHz under external modulation with −7.4-dB optical feedback. The feedback insensitivity results from the low linewidth enhancement factor, the high damping, the absence of off-resonance emission states, and the shorter carrier lifetime. This letter paves the way for future on chip high-speed integrated circuits operating without optical isolators.