Search

Code-switching speech refers to a means of expression by mixing two or more languages within a single utterance. Automatic Speech Recognition (ASR) with End-to-End (E2E) modeling for such speech can be a challenging task due to the lack of data. In this study, we investigate text generation and injection for improving the performance of an industry commonly-used streaming model, Transformer-Transducer (T-T), in Mandarin-English code-switching speech recognition. We first propose a strategy to generate code-switching text data and then investigate injecting generated text into T-T model explicitly by Text-To-Speech (TTS) conversion or implicitly by tying speech and text latent spaces. Experimental results on the T-T model trained with a dataset containing 1,800 hours of real Mandarin-English code-switched speech show that our approaches to inject generated code-switching text significantly boost the performance of T-T models, i.e., 16% relative Token-based Error Rate (TER) reduction averaged on three evaluation sets, and the approach of tying speech and text latent spaces is superior to that of TTS conversion on the evaluation set which contains more homogeneous data with the training set., Comment: Accepted by ICASSP 2023

A key problem in spoken language identification (LID) is how to design effective representations which are specific to language information. Recent advances in deep neural networks have led to significant improvements in results, with deep end-to-end methods proving effective. This paper proposes a novel network which aims to model an effective representation for high (first and second)-order statistics of LID-senones, defined as being LID analogues of senones in speech recognition. The high-order information extracted through bilinear pooling is robust to speakers, channels and background noise. Evaluation with NIST LRE 2009 shows improved performance compared to current state-of-the-art DBF/i-vector systems, achieving over 33% and 20% relative equal error rate (EER) improvement for 3s and 10s utterances and over 40% relative Cavg improvement for all durations.

The energy function method developed so far assumes that the active load is constant. However, most of the actual load is voltage dependent, which hinders the application of the energy analysis method to the actual power system. On the other hand, the fast development of measurement techniques, especially the wide area measurement systems (WAMS) applied widely in worldwide power system nowadays, presses for the on-line transient stability analysis tools based on the system measurements. The voltage-dependent load into the energy analysis of power system transient stability is discussed. The effect of the comprehensive load characteristics on the system energy is discussed and a WAMS-based energy indicator on the system stability status is proposed. The indicator is easy to be calculated and thus real-time in the operation centre is monitored. The proposed method does not rely on any specific load model, and various load characteristics on different buses can be considered. Simulations on two test systems verify the effectiveness of the proposed method.

Load modeling is very important to power system operation and control. Measurement-based load modeling has been widely practiced in recent years. Mathematically, measurement-based load modeling problem are closely related to the parameter identification area. Consequently, an efficient optimization method is needed to derive the load model parameters based on the feedback of estimation errors between the measurements and model outputs. This paper reports our work on applying genetic algorithms on measurement-based load modeling research. Due to its robustness to the initial guesses on the load model parameters, genetic algorithms are very suitable for load model parameter identification. Two cases including both the real measurement in a power station and the digital simulation are studied in the paper. For comparison purpose, the classical nonlinear least square estimation method is also applied to find the load model parameters. The simulated outputs from the load model confirm the efficiency of genetic algorithms in measurement-based load modeling analysis. Future work will focus on fastening the converging speed of the genetic algorithms, and/or utilizing more efficient evolutionary computation methods.

Load modeling is very important for power system dynamic analysis and control. The composite load model widely applied recently in the power system operation centre consists of the static load and the equivalent motor. Current practices in measurement-based load modeling identify all the parameters in this composite load model. However, it is not clear whether all these parameters could be identified from the measurements. This paper investigates the identifiability of the equivalent motor parameters in the composite load model. Trajectory sensitivity approach is applied first to find the motor parameters that have great effects on the measured active as well as reactive load dynamics. The analysis results show that the motor outputs have various sensitivities with respect to the parameters. Since the voltage disturbance, the active load and the reactive load dynamics are applied to identify the motor parameters, those parameters affecting the measurements to a great extent are observable, thus identifiable from the measurements; while those that have little effects on the motor outputs are unobservable from the measurements and consequently unidentifiable. The case studies verify the identifiability of the motor parameters.

The accuracy of the load model has great effects on power system stability analysis and control. Based on our practice in China on modeling load from field measurements, this paper systematically develops a measurement-based composite load model. Principles guiding the load modeling practice are discussed based on detailed analysis on stochastic characteristics of the modeling procedure. The structure of the measurement-based composite load model is presented. A multicurve identification technique is described to derive parameters. The generalization capability of this built load model is also investigated in this paper. Two cases are studied to illustrate the accuracy of the developed load model on describing the load dynamic characteristics in the actual power system.

The representation of load dynamic characteristics remains an area of great uncertainty and has become a limiting factor for power system analysis and control. The random nature of the load makes load modeling a very difficult problem, which becomes even