High quality voice conversion using prosodic and high-resolution spectral features

Voice conversion methods have advanced rapidly over the last decade. Studies have shown that speaker characteristics are captured by spectral feature as well as various prosodic features. Most existing conversion methods focus on the spectral feature as it directly represents the timbre characteristics, while some conversion methods have focused only on the prosodic feature represented by the fundamental frequency. In this paper, a comprehensive framework using deep neural networks to convert both timbre and prosodic features is proposed. The timbre feature is represented by a high-resolution spectral feature. The prosodic features include F0, intensity and duration. It is well known that DNN is useful as a tool to model high-dimensional features. In this work, we show that DNN initialized by our proposed autoencoder pretraining yields good quality DNN conversion models. This pretraining is tailor-made for voice conversion and leverages on autoencoder to capture the generic spectral shape of source speech. Additionally, our framework uses segmental DNN models to capture the evolution of the prosodic features over time. To reconstruct the converted speech, the spectral feature produced by the DNN model is combined with the three prosodic features produced by the DNN segmental models. Our experimental results show that the application of both prosodic and high-resolution spectral features leads to quality converted speech as measured by objective evaluation and subjective listening tests.     

Corpus based part-of-speech tagging

In natural language processing, a crucial subsystem in a wide range of applications is a part-of-speech (POS) tagger, which labels (or classifies) unannotated words of natural language with POS labels corresponding to categories such as noun, verb or adjective. Mainstream approaches are generally corpus-based: a POS tagger learns from a corpus of pre-annotated data how to correctly tag unlabeled data. Presented here is a brief state-of-the-art account on POS tagging. POS tagging approaches make use of labeled corpus to train computational trained models. Several typical models of three kings of tagging are introduced in this article: rule-based tagging, statistical approaches and evolution algorithms. The advantages and the pitfalls of each typical tagging are discussed and analyzed. Some rule-based and stochastic methods have been successfully achieved accuracies of 93–96 %, while that of some evolution algorithms are about 96–97 %.     

Chaotic feature analysis and forecasting of Liujiang River runoff

Because most of runoff time series with limited amount of data reveal inherently nonlinear and stochastic characteristics and tend to show chaotic behavior, strategies based on chaotic analysis are popular methods to analyze them from real systems in nonlinear dynamics. Only one kind of predicted method for yearly rainfall-runoff forecasting cannot achieve perfect performance. Thus, a mixture strategy denoted by WT-PSR-GA-NN, which is composed of wavelet transform (WT), phase space reconstruction (PSR), neural network (NN) and genetic algorithm (GA), is presented in this paper. In the WT-PSR-GA-NN framework, the process to deal with time series gathered from Liujiang River runoff data is given as follows: (1) the runoff time series was first decomposed into low-frequency and high-frequency sub-series by wavelet transformation; (2) the two sub-series were separately and independently reconstructed into phase spaces; (3) the transformed time series in the reconstructed phase spaces were modeled by neural network, which is trained by genetic algorithm to avoid trapping into local minima; (4) the predicted results in low-frequency parts were combined with the ones of high-frequency parts, and reconstructed with wavelet inverse transformation, to form the future behavior of the runoff. Experiments show that WT-PSR-GA-NN is effective and its forecasting results are high in accuracy not only for the short-term yearly hydrological time series but also for the long-term one. The comparison results revealed that the overall forecasting performance of WT-PSR-GA-NN proposed by us is superior to other popularity methods for all the test cases. We can conclude that WT-PSR-GA-NN can not only increase the forecasted accuracy, but also its own competitiveness in efficiency, effectiveness and robustness.     

Adaptive Tracking Control for Nonlinear Systems with a Class of Input Nonlinearities

An adaptive tracking control approach is presented for nonlinear systems with a class of input nonlinearities. A generalized model has been developed for a class of non‐smooth nonlinearities that include dead‐zone, backlash and ‘backlash‐like’ hysteresis. By using the developed model and Nussbaum‐gain technique, the problem of input nonlinearity is solved perfectly. The proposed method is available even when the designer is uncertain about the type of input nonlinearities mentioned above, and the knowledge on the bounds of these nonlinearity parameters is not required. Furthermore, it is proved that all closed‐loop signals are bounded and the tracking error converges to a small residual set asymptotically. Two simulation examples are provided to demonstrate the effectiveness of the proposed method.     

Teleportation of a general two-photon state employing a polarization-entangled $$\chi $$ state with nondemolition parity analyses

Employing a polarization-entangled \(\chi \) state, which is a four-photon genuine entangled state, we propose a protocol teleporting a general two-photon polarization state. Firstly, the sender needs to perform one Controlled-NOT gate, one Hadamard gate, and one Controlled-NOT gate on the state to be teleported in succession. Secondly, the sender performs local nondemolition parity analyses based on cross-Kerr nonlinearities and publicizes the achieved outcomes. Finally, conditioned on the sender’s analysis outcomes, the receiver executes the single-photon unitary transformation operations on his own photons to obtain the state originally sit in the sender’s location. Due to the employment of nondemolition parity analyses rather than four-qubit joint measurement, it can be realized more feasible with currently available technologies. Moreover, the resources of Bell states can be achieved because the nondestructive measurement is exploited, which facilitates other potential tasks of quantum information processing.     

Measurement device-independent quantum key distribution with heralded pair coherent state

The original measurement device-independent quantum key distribution is reviewed, and a modified protocol using heralded pair coherent state (HPCS) is proposed to overcome the quantum bit error rate associated with the dark count rate of the detectors in long-distance quantum key distribution. Our simulation indicates that the secure transmission distance can be improved evidently with HPCS owing to the lower probability of vacuum events when compared with weak coherent source scenario, while the secure key rate can be increased with HPCS due to the higher probability of single-photon events when compared with heralded single-photon source scenario. Furthermore, we apply the finite key analysis to the decoy state MDI-QKD with HPCS and obtain a practical key rate.     

state estimation with fading measurements,randomly varying nonlinearities and probabilistic distributed delays

In this paper, the state estimation problem is investigated for a class of discrete‐time stochastic systems in simultaneous presence of three network‐induced phenomena, namely, fading measurements, randomly varying nonlinearities and probabilistic distributed delays. The channel fading is characterized by the ?th‐order Rice fading model whose coefficients are mutually independent random variables with given probability density functions. Two sequences of random variables obeying the Bernoulli distribution are utilized to govern the randomly varying nonlinearities and probabilistic distributed delays. The purpose of the problem addressed is to design an state estimator such that the dynamics of the estimation errors is stochastically stable and the prespecified disturbance rejection attenuation level is guaranteed. Through intensive stochastic analysis, sufficient conditions are established under which the addressed state estimation problem is recast as a convex optimization one that can be solved via the semi‐definite program method. Finally, a simulation example is provided to show the usefulness of the proposed state estimation scheme. Copyright © 2014 John Wiley & Sons, Ltd.     

Probability‐weighted nonlinear stochastic optimal control strategy of quasi‐integrable Hamiltonian systems with uncertain parameters

The nonlinear stochastic optimal control problem of quasi‐integrable Hamiltonian systems with uncertain parameters is investigated. The uncertain parameters are described by using a random vector with λ probability density function. First, the partially averaged Itô stochastic differential equations are derived by using the stochastic averaging method for quasi‐integrable Hamiltonian systems. Then, the dynamical programming equation is established based on stochastic dynamical programming principle. By minimizing the dynamical programming equation with respect to control forces, the optimal control forces can be derived, which are functions of the uncertain parameters. The final optimal control forces are then determined by probability‐weighted average of the obtained control forces with the probability density of the uncertain parameters as weighting function. The mean control effectiveness and mean control efficiency are used to evaluate the proposed control strategy. The robustness of the proposed control is measured by using the ratios of the variation coefficients of mean control effectiveness and mean control efficiency to the variation coefficients of uncertain parameters. Finally, two examples are given to illustrate the proposed control strategy and its effectiveness and robustness. Copyright © 2014 John Wiley & Sons, Ltd.