Search

This paper proposes a new strategy for moving target detection and localization based on monocular vision. Firstly, to detect a moving target with large displacement and high speed accurately, two consecutive video images captured by a monocular camera are preprocessed using the enhancement and denoising methods. Then, the optical flow representing motion information is calculated iteratively by the modified Lucas-Kanade optical flow method. Secondly, a new interest region extraction method is developed to overcome the negative impacts caused by the noises in the background. Specifically, this proposed method combines a two-level image segmentation strategy from coarse to fine, including median filtering, two-direction dynamic threshold segmentation, the Otsu method, and morphological processing. Thirdly, a low computational cost target localization algorithm is proposed based on pin-hole imaging theory. Besides, it only uses two-dimensional image and camera parameters to obtain the moving target's position in the three-dimensional space. Finally, experimental results show that the proposed strategy can effectively eliminate noise interferences and realize moving target detection, extraction, and localization.

In multi-talker cases, overlapped speech degrades the speaker verification (SV) performance dramatically. To tackle this challenging problem, speech separation with multi-channel techniques can be adopted to extract each speaker’s signals to improve the SV performance. In this paper, a joint training framework of the front-end multi-look speech separator and the back-end speaker embedding extractor is proposed for multi-channel overlapped speech. To better leverage the complementarity between the speech separator and the speaker embedding extractor, several training strategies are proposed to jointly optimize the two modules. Experimental results show that the proposed joint training framework significantly outperforms the individual SV system by around 52% relative EER reduction. Additionally, the robustness of the proposed framework is further evaluated under different conditions.

Non-autoregressive (NAR) transformer models have achieved significantly inference speedup but at the cost of inferior accuracy compared to autoregressive (AR) models in automatic speech recognition (ASR). Most of the NAR transformers take a fixed-length sequence filled with MASK tokens or a redundant sequence copied from encoder states as decoder input, they cannot provide efficient target-side information thus leading to accuracy degradation. To address this problem, we propose a CTC-enhanced NAR transformer, which generates target sequence by refining predictions of the CTC module. Experimental results show that our method outperforms all previous NAR counterparts and achieves 50x faster decoding speed than a strong AR baseline with only 0.0 ~ 0.3 absolute CER degradation on Aishell-1 and Aishell-2 datasets., Comment: Accepted to ICASSP 2021, final version

The form of the original stochastic gradient descent (SGD) algorithm accords with the definition of forward Euler method, which has some inherent defects. Thus, in order to improve the original SGD algorithm, a Lagrange-type 1-step-ahead numerical differentiation method based parameter update algorithm is presented and validated. Instinctively, the new algorithm can fix some inherent flaws of the SGD algorithm. However, a series of experimental results show that the Lagrange-type 1-step-ahead numerical differentiation method cannot be applied to reduce the computational error of SGD. In addition, this method makes the model appear the phenomenon of non-convergence. Finally, on the basis of comparative experiments, the divergence phenomenon is analyzed and explained.

It is necessary to make physical constraints on the joints for the redundant robot motion control in order to avoid damage. In this paper, a discrete-time neural network model with minimum kinetic energy as the performance index is proposed, which has predominant convergence performance. Then, a solution in robot motion control is studied and further transformed into a dynamic quadratic programming (QP) with equality and inequality constraints. In addition, for solving the formulated QP problem, a continuous-time neural network model is designed by introducing the Lagrange multiplier method, and a discrete-time neural network model is obtained by the Euler forward difference formula. Moreover, the simulations on robot motion control are carried out, and the simulative results further substantiate the superiority, thus extending a solution for motion control of redundant robots with double-bound constraints.

State-of-the-art singing voice synthesis (SVS) models can generate natural singing voice of a target speaker, given his/her speaking/singing data in the same language. However, there may be challenging conditions where only speech data in a non-target language of the target speaker is available. In this paper, we present a cross-lingual SVS system that can synthesize an English speaker’s singing voice in Mandarin from musical scores with only her speech data in English. The pro-posed cross-lingual SVS system contains four parts: a BLSTM based duration model, a pitch model, a cross-lingual acoustic model and a neural vocoder. The acoustic model employs encoder-decoder architecture conditioned on pitch, phoneme duration, speaker information and language information. An adversarially-trained speaker classifier is employed to discourage the text encodings from capturing speaker information. Objective evaluation and subjective listening tests demonstrate that the proposed cross-lingual SVS system can generate singing voice with decent naturalness and fair speaker similarity. We also find that adding singing data or multi-speaker monolingual speech data further improves generalization on pronunciation and pitch accuracy.

In this paper, a fault feature extraction algorithm model of distribution network based on wavelet entropy is designed, and combined with a large number of simulation data and field data for test and analysis, to improve the accuracy of line fault diagnosis and troubleshooting. The case verification results show that the feature extraction algorithm and feature gene library based on wavelet entropy can effectively identify the normal state, three-phase short circuit fault, two-phase short-circuit fault, two-phase grounding fault and single-phase grounding fault of distribution network transient signals, which meets the requirements of sustainable development of distribution network, and has good economic and social benefits.

Synthesizing fluent code-switched (CS) speech with consistent voice using only monolingual corpora is still a challenging task, since language alternation seldom occurs during training and the speaker identity is directly correlated with language. In this paper, we present a bilingual phonetic posteriorgram (PPG) based CS speech synthesizer using only monolingual corpora. The bilingual PPG is used to bridge across speakers and languages, which is formed by stacking two monolingual PPGs extracted from two monolingual speaker-independent speech recognition systems. It is assumed that bilingual PPG can represent the articulation of speech sounds speaker-independently and captures accurate phonetic information of both languages in the same feature space. The proposed model first extracts bilingual PPGs from training data. Then an encoder- decoder based model is used to learn the relationship between input text and bilingual PPGs, and the bilingual PPGs are mapped to acoustic features using bidirectional long-short term memory based model conditioned on speaker embedding to control speaker identity. Experiments validate the effectiveness of the proposed model in terms of speech intelligibility, audio fidelity and speaker consistency of the generated code-switched speech.

Hand-crafted spatial features (e.g., inter-channel phase difference, IPD) play a fundamental role in recent deep learning based multi-channel speech separation (MCSS) methods. However, these manually designed spatial features are hard to incorporate into the end-to-end optimized MCSS framework. In this work, we propose an integrated architecture for learning spatial features directly from the multi-channel speech waveforms within an end-to-end speech separation framework. In this architecture, time-domain filters spanning signal channels are trained to perform adaptive spatial filtering. These filters are implemented by a 2d convolution (conv2d) layer and their parameters are optimized using a speech separation objective function in a purely data-driven fashion. Furthermore, inspired by the IPD formulation, we design a conv2d kernel to compute the inter-channel convolution differences (ICDs), which are expected to provide the spatial cues that help to distinguish the directional sources. Evaluation results on simulated multi-channel reverberant WSJ0 2-mix dataset demonstrate that our proposed ICD based MCSS model improves the overall signal-to-distortion ratio by 10.4% over the IPD based MCSS model., Comment: accepted in ICASSP 2020

While question answering (QA) systems have witnessed great breakthroughs in reading comprehension (RC) tasks, spoken question answering (SQA) is still a much less investigated area. Previous work shows that existing SQA systems are limited by catastrophic impact of automatic speech recognition (ASR) errors [1] and the lack of large-scale real SQA datasets [2]. In this paper, we propose using contextualized word representations to mitigate the effects of ASR errors and pretraining on existing textual QA datasets to mitigate the data scarcity issue. New state-of-the-art results have been achieved using contextualized word representations on both the artificially synthesised and real SQA benchmark data sets, with 21.5 EM/18.96 F1 score improvement over the sub-word unit based baseline on the Spoken-SQuAD [1] data, and 13.11 EM/10.99 F1 score improvement on the ODSQA data [2]. By further fine-tuning pre-trained models with existing large scaled textual QA data, we obtained 38.12 EM/34.1 F1 improvement over the baseline of fine-tuned only on small sized real SQA data.

Neural network language model (NNLM) is an essential component of industrial ASR systems. One important challenge of training an NNLM is to leverage between scaling the learning process and handling big data. Conventional approaches such as block momentum provides a blockwise model update filtering (BMUF) process and achieves almost linear speedups with no performance degradation for speech recognition. However, it needs to calculate the model average from all computing nodes (e.g., GPUs) and when the number of computing nodes is large, the learning suffers from the severe communication latency. As a consequence, BMUF is not suitable under restricted network conditions. In this paper, we present a decentralized BMUF process, in which the model is split into different components, each of which is updated by communicating to some randomly chosen neighbor nodes with the same component, followed by a BMUF-like process. We apply this method to several LSTM language modeling tasks. Experimental results show that our approach achieves consistently better performance than conventional BMUF. In particular, we obtain a lower perplexity than the single-GPU baseline on the wiki-text-103 benchmark using 4 GPUs. In addition, no performance degradation is observed when scaling to 8 and 16 GPUs., This paper is accepted in the technical program in ICASSP 2020. Session FR1.L3: Language Modeling, Location Room 118-119, Presentation Lecture,Presentation Time: Friday, 08 May, 09:40 - 10:00, Topic Human Language Technology:[HLT-LANG] Language Modeling

Adapting speaker verification (SV) systems to a new environment is a very challenging task. Current adaptation methods in SV mainly focus on the backend, i.e, adaptation is carried out after the speaker embeddings have been created. In this paper, we present a DNN-based adaptation method using maximum mean discrepancy (MMD). Our method exploits two important aspects neglected by previous research. First, instead of minimizing domain discrepancy at utterance-level alone, our method minimizes domain discrepancy at both frame-level and utterance-level, which we believe will make the adaptation more robust to the duration discrepancy between training data and test data. Second, we introduce a consistency regularization for unlabelled target-domain data. The consistency regularization encourages the target speaker embeddings robust to adverse perturbations. Experiments on NIST SRE 2016 and 2018 show that our DNN adaptation works significantly better than the previously proposed DNN adaptation methods. What’s more, our method works well with backend adaptation. By combining the proposed method with backend adaptation, we achieve a 9% improvement over backend adaptation in SRE18.

Deep-learning based speech separation models confront poor generalization problem that even the state-of-the-art models could abruptly fail when evaluating them in mismatch conditions. To address this problem, we propose an easy-to-implement yet effective consistency based semi-supervised learning (SSL) approach, namely Mixup-Breakdown training (MBT). It learns a teacher model to "breakdown" unlabeled inputs, and the estimated separations are interpolated to produce more useful pseudo "mixup" input-output pairs, on which the consistency regularization could apply for learning a student model. In our experiment, we evaluate MBT under various conditions with ascending degrees of mismatch, including unseen interfering speech, noise, and music, and compare MBT's generalization capability against state-of-the-art supervised learning and SSL approaches. The result indicates that MBT significantly outperforms several strong baselines with up to 13.77% relative SI-SNRi improvement. Moreover, MBT only adds negligible computational overhead to standard training schemes., Accepted in a Lesson session in ICASSP2020

Keyword spotting (KWS) is in great demand in smart devices in the era of Internet of Things. Albeit recent progresses, the performance of KWS, measured in false alarms and false rejects, may still degrade significantly under the far field and noisy conditions. In this paper, we propose integrating multiple beamformed signals and a microphone signal as input to an end-to-end KWS model and leveraging the attention mechanism to dynamically tune the model’s attention to the reliable input sources. We demonstrate, on our large simulated and recorded noisy and far-field evaluation sets, that our proposed approach significantly improves the KWS performance and reduces the computation cost against the baseline KWS systems.

Techniques for accent conversion (AC) aim to convert non-native to native accented speech. Conventional AC methods try to convert only the speaker identity of a native speaker’s voice to that of the non-native accented target speaker, leaving the underlying content and pronunciations unchanged. This hinders their practical use in real-world applications, because native-accented utterances are required at conversion stage. In this paper, we present an end-to-end framework, which is able to conduct AC from non-native-accented utterances without using any native-accented utterances during online conversion. We achieve this by independently extracting linguistic and speaker representations from non-native accented speech and condition a speech synthesis model on these representations to generate native-accented speech. Experiments on open-source data corpora show that the proposed system can convert Hindi-accented English speech into native American English speech with high naturalness, which is indistinguishable from native-accented recordings in terms of accent.

Deep learning based speech separation approaches have received great interest, among which the recent speaker-aware speech enhancement methods are promising for solving difficulties such as arbitrary source permutation and unknown number of sources. In this paper, we propose a novel training framework which jointly learns the speaker-conditioned target speaker extraction model and its associated speaker embedding model. The resulting unified model directly learns the appropriate speaker embedding for improved target speech enhancement. We demonstrate, on our large simulated noisy and far-field evaluation sets of overlapped speech signals, that our proposed approach significantly improves the speech enhancement performance compared to the baseline speaker-aware speech enhancement models.

Constant frequency double rotor generator (CFDRG) is a novel doubly-fed machine for wind power generation with auxiliary excitation of permanent magnets (PMs). This paper proposes a grid connected wind power generation system employing the CFDRG. The operation of the PM rotor is very crucial to the dynamics of the machine. So the operation under certain disturbances and stability analysis of the CFDRG are carried out to find the safe operating limit of the machine. Notably, the CFDRG under wind gust and grid disturbances are analyzed in aims to find the dynamic performance of the power generation system. Theoretical analysis and simulations in MATLAB have validated the effectiveness of the proposed wind power generation system.

In this paper, we present a speech enhancement framework that leverages phonetic information obtained from the acoustic model. It consists of two separate components: (i) a long short-term memory recurrent neural network (LSTM-RNN) based speech enhancement model that takes the combination of log-power spectra (LPS) and phonetic embedding features as input to predict the complex ideal ratio mask (cIRM); and (ii) a convolutional, long short-term memory and fully connected deep neural network (CLDNN) based acoustic model that extracts the phonetic feature vector in the hidden units of its LSTM layer. Our experimental results show that the proposed framework outperforms both the conventional and phoneme-dependent speech enhancement systems under various noisy conditions, generalizes well to unseen conditions, and performs robustly to the speech interference. We further demonstrate its superior enhancement performance on unvoiced speech and report a preliminary yet promising recognition experiment on real test data.

This study proposes a novel scheme of syllable-dependent discriminative speaker embedding learning for small footprint text-dependent speaker verification systems. To suppress undesired syllable variation and enhance the power of discrimination inherited in the frame-level features, we design a novel syllable-dependent clustering loss to optimize the network. Specifically, this loss function utilizes syllable labels as auxiliary supervision information to explicitly maximize inter-syllable divisibility and intra-syllable compactness between the learned frame-level features. Successively, we propose two syllable-dependent pooling mechanisms to aggregate the frame-level features to several syllable-level features by averaging those features corresponding to each syllable. The utterance-level speaker embeddings with powerful discrimination are then obtained by concatenating the syllable-level features. Experimental results on Tencent voice wake-up dataset show that our proposed scheme can accelerate the network convergence and achieve significant performance improvement against the state-of-the-art methods.

In this paper, a set of wireless communication module based on LoRa technology is developed, and a complete local area network wireless communication solutions satisfying long distance and low power consumption is proposed, which has wide wireless coverage, strong penetration and star architecture. By using LoRa communication, the data of the distribution terminal can be transmitted to the data collection unit seamlessly. The data is more safe and reliable. Through practical field application, the communication system is verified.

Hyperlens consisted by period plasmonic metamaterials is an interesting technology for the optical imaging of arbitrary objects beyond the diffraction limit in the far field, which has great potential applications in nanophotonics. However, the functionality of previously presented hyperlens is limited to image two-dimensional (2-D) objects and work at a certain wavelength due to the dispersion of metal, dramatically impose restrictions on its practical application. In this paper, we proposed and theoretically demonstrated a multiwavelength tunable hyperlens, with the ability of 3-D imaging by focal plane tuning achieved by electro-optical modulation. The simulation results indicate that full-width half-maximum (FWHM) of minimum imaging spot can be reduced down to the level of less than half-wavelength of the incident in a wide wavelength range real-timely, down to the level of ∼200 nm using visible light illumination. In addition, the focal position of the hyperlens can also be tuned in large spatial scope, which greatly facilitates the application of the hyperlens in optical signal acquisition and processing.

The speech corpus with labeled prosodic structure information is crucial for text-to-speech (TTS) synthesis to train a reliable model that can generate high quality natural synthetic speech. Traditional manual prosodic structure labeling is laborious and time-consuming and may encounter an inconsistency problem caused by different annotators. Automatic prosodic labeling is thus desirable, which can not only speed up the labeling process, but also protect the labeling results from the inconsistency problem. This paper presents a DNN-BGRU-CRF hybrid neural network, which aggregates the advantages of deep neural network, bidirectional gated recurrent units and conditional random fields, to label three-level prosodic structure boundaries. It exploits both text and acoustic cues in a neural network framework. Experimental results demonstrate the effectiveness of the proposed model.

In the context of mobile gaze tracking techniques, a 3D gaze point can be calculated as the middle point between two 3D visual axes. To infer gaze directions and eyeball positions, a nonlinear optimization problem is typically formulated to minimize the angular disparities between the training gaze directions and prediction ones. Nonetheless, the experimental results reported by some previous works show that this kind of approaches are very likely to yield large prediction errors hence considered less useful for human-machine interactions. In this study, we aim to address this widespread issue in three aspects. At first, instead of using a global regression model, a simple local polynomial model is proposed to back-project a pupil center onto its corresponding visual axis. Based on the Leave-One-Out cross-validation criterion, the partition structure is automatically learned in the process of resolving a homography-like relationship. Secondly, a good starting point for nonlinear-optimization is obtained by the image eyeball center, which can be estimated by systematic parallax errors. Meanwhile, it is necessary to add the suitable constraints for 3D eye positions. Otherwise, the optimization may end up with trivial solutions, i.e., faraway eye positions. Thirdly, we explore a strategy for designing the spatial distribution of calibration points in a principled manner. The experiment results demonstrate that an encouraging gaze estimation accuracy can be achieved by our proposed framework for both the normal vision and eyewear users.

Infrared imaging spectrometer (IRIS) often suffers from overlapped bands and random noises, which limit the precision of subsequent processing in robot vision sensing. To address this problem, we propose a novel Gabor transform-based infrared spectrum restoration method by successfully exploring the intrinsic structure of the clean IR spectrum from the degraded one. At first, a total variation (TV) regularized Gabor coefficients adjustment descriptor is designed and incorporated into the spectrum restoration model. Then, the proposed model is inferred via an efficient optimization approach based on split Bregman iteration method. Comprehensive experiments illustrate the significant and consistent improvements of the developed model over state-of-the-art approaches. The restored high-resolution spectrum can be utilized for detecting the different materials in the robot visual tracking systems.

Prosodic structure prediction is a key part of the text analysis front-end of the text-to-speech (TTS) system. It predicts prosodic boundary tags given the input text context, which is essential to the naturalness of synthesized speech. Conventional methods such as conditional random fields (CRF) and recurrent neural network (RNN) have been successfully applied to this task. However, the lack of modeling temporal dependencies at different scopes (the short-term dependency as well as the long-span dependency across the entire sentence) limits their performance. In this paper, we propose a self-attention network with semantic features extracted by a pre-trained bidirectional encoder representations from Transformers (BERT) model to predict the prosodic structure. Experimental results show that the proposed approach outperforms the strong baseline CRF model with an absolute improvement of 3.4% in total accuracy.

A poly-L-lysine assisted self-assembly method based on the surface chemistry of silica microspheres and the textured polycrystalline silicon solar cell surface is proposed in this article. With four-step solution-based processes, silica microspheres are deposited tightly on the surface of the device, acting as an anti-reflection layer. By tuning the assembly time, uniform distributions of silica microspheres with varying density can be achieved. Furthermore, the electrical tests show a maximum relative efficiency enhancement of 9.35% with solar cell area of 2 cm * 2 cm under AM1.5 illumination. Optical micrographs and reflection spectra are used to further characterize the light-trapping properties of silica microspheres. This strategy may pave the way for large-area and cost-effective assembly of silica spheres or silica coated nanoparticles on different substrates for solar energy harvesting applications.

Deep neural network based speaker embeddings have attracted much attention in text-independent speaker verification task. In addition to the network architecture, an appropriate design of the loss function is crucial for the deep discriminative embedding extractor. Inspired by the success of Large Margin Cosine Loss (LMCL) in face recognition, we propose an enhanced LMCL named boundary discriminative LMCL (BD-LMCL) to emphasize the discriminative information inherited in the speaker boundaries. Unlike LMCL, where all training samples contribute equally for the objective function, only the samples around the speaker boundaries are considered during the network training with BD-LMCL. Specifically, those samples close to the boundaries are dynamically selected using top-k zero-one loss. Experimental results on a short duration corpus Android Cellphone and NIST SRE 2012 demonstrate better performance compared to LMCL and other popular loss functions.

Code-switching is a common phenomenon in many multilingual communities and presents a challenge to automatic speech recognition (ASR). In this paper, three approaches are investigated to improve end-to-end speech recognition on Mandarin-English code-switching task. First, multi-task learning (MTL) is introduced which enables the language identity information to facilitate Mandarin-English code-switching ASR. Second, we explore wordpieces, as opposed to graphemes, as English modeling units to reduce the mod-eling unit gap between Mandarin and English. Third, we employ transfer learning to utilize larger amount of monolingual Mandarin and English data to compensate the data sparsity issue of a code-switching task. Significant improvements are observed from all three approaches. With all three approaches combined, the final system achieves a character error rate (CER) of 6.49% on a real Mandarin-English code-switching task.

Recently, attention-based end-to-end automatic speech recognition system (ASR) has shown promising results. One of the limitations of an attention-based ASR system is that its language model (LM) component has to be implicitly learned from transcribed speech data which prevents one from uti-lizing plenty of text corpora to improve language modeling. In this work, the Component Fusion method is proposed to incorporate externally trained neural network (NN) LM into an attention-based ASR system. During training stage we equip the attention-based system with an additional LM component which is replaced by an externally trained NN LM at decoding stage. Experimental results show that the proposed Component Fusion outperforms two prior LM fusion approaches, i.e., Shallow Fusion and Cold Fusion, in both out-of-domain and in-domain scenarios. Further improvements can be achieved when combining Component and Shallow Fusion.

In this paper, we present a joint training framework between the multi-channel beamformer and the acoustic model for noise robust automatic speech recognition (ASR). The complex ratio mask (CRM), demonstrated to be more effective than the ideal ratio mask (IRM), is proposed to estimate the covariance matrix for the beamformer. Minimum Variance Distortionless Response (MVDR) beamformer and Generalized Eigenvalue (GEV) beamformer are both investigated under the CRM-based joint training architecture. We also propose a robust mask pooling strategy among multiple channels. A long short-term memory (LSTM) based language model is utilized to re-score hypotheses which further improves the overall performance. We evaluate the proposed methods on CHiME-4 challenge dataset. The CRM based system achieves a relative 10% reduction on word error rate (WER) compared with the IRM based system. Without sequence discriminative training, our best single system already achieves an average WER 2.72% on the test set which is comparable to the state-of-the-art.

Recurrent neural networks, such as gated recurrent units (GRUs) and long short-term memory (LSTM), are widely used on acoustic modeling for speech synthesis. However, such sequential generating processes are not friendly to today’s massively parallel computing devices. We introduce a fully convolutional neural network (CNN) model, which can effiently run on parallel processers, for speech synthesis. To improve the quality of the generated acoustic features, we strengthen our model with variational inference. We also use quasi-recurrent neural networks (QRNNs) to smoothen the generated acoustic features. Finally, a high-quality parallel WaveNet model is used to generate audio samples. Our contributions are twofold. First, we show that CNNs with variational inference can generate highly natural speech on a par with end-to-end models; the use of QRNNs further improves the synthetic quality by reducing trembling of generated acoustic features and introduces very little run-time overheads. Second, we show some techniques to further speed up the sampling process of the parallel WaveNet model.

The recent exploration of deep learning for supervised speech separation has significantly accelerated the progress on the multi-talker speech separation problem. Multi-channel extension has attracted much research attention due to the benefit of spatial information in far-field acoustic environments. In this paper, We review the most recent models of multi-channel permutation invariant training (PIT), investigate spatial features formed by microphone pairs and their underlying impact and issue, present a multi-band architecture for effective feature encoding, and conduct a model integration between single-channel and multi-channel PIT for resolving the spatial overlapping problem in the conventional multi-channel PIT framework. The evaluation confirms the significant improvement achieved with the proposed model and training approach for the multi-channel speech separation.

Acoustic model and language model (LM) have been two major components in conventional speech recognition systems. They are normally trained independently, but recently there has been a trend to optimize both components simultaneously in a unified end-to-end (E2E) framework. However, the performance gap between the E2E systems and the traditional hybrid systems suggests that some knowledge has not yet been fully utilized in the new framework. An observation is that the current attention-based E2E systems could produce better recognition results when decoded with LMs which are independently trained with the same resource.In this paper, we focus on how to improve attention-based E2E systems without increasing model complexity or resorting to extra data. A novel training strategy is proposed for multi-task training with the connectionist temporal classification (CTC) loss. The sequence-based minimum Bayes risk (MBR) loss is also investigated. Our experiments on SWB 300hrs showed that both loss functions could significantly improve the baseline model performance. The additional gain from joint-LM decoding remains the same for CTC trained model but is only marginal for MBR trained model. This implies that while CTC loss function is able to capture more acoustic knowledge, MBR loss function exploits more word/character dependency.

Audio super-resolution is the task to increase the sampling rate of a given low-resolution (i.e. low sampling rate) audio. One of the most popular approaches for audio super-resolution is to minimize the squared Euclidean distance between the reconstructed signal and the high sampling rate signal in a point-wise manner. However, such approach has intrinsic limitations, such as the regression to mean problem. In this work, we introduce a novel auto-regressive method for the speech super-resolution task, which utilizes WaveNet to model the distribution of the target high-resolution signal conditioned on the log-scale mel-spectrogram of the low-resolution signal. As an auto-regressive neural network, WaveNet uses the negative log-likelihood as the objective function, which is much more suitable for highly stochastic process such as speech waveform, instead of the Euclidean distance. We also train a parallel WaveNet to speed up the generating process to real-time. In the experiments, we perform speech super-resolution by increasing the sampling rate from 4kHz to 16kHz on the VCTK corpus. The proposed method can achieve the improvement of ∼2 dB over the base-line deep residual convolutional neural network (CNN) under the Log-Spectral Distance (LSD) metric.

Convolutional neural networks have achieved wide success in RGB saliency detection. Recently, the advent of RGB-D sensors such as Kinect provide additional geometric saliency cues. However, the key challenge for RGB-D salient object detection that how to fuse RGB and depth information sufficiently is still under-studied. Traditional works mainly follow the two-stream architecture and combine RGB and depth features/decisions in an early or late point. The multi-modal fusion stage is performed by directly concatenating the features from two modalities without selection. In this work, we address this question by proposing a novel network with a distinguished insight: A selection module is significantly helpful for more informative and sufficient cross-modal cross-level combination. To this end, we introduce a top-down RGB-D fusion network which integrates an attention-aware cross-modal cross-level fusion block in each level to select discriminative features from each level and each modality. Extensive experiments on public datasets show that the proposed network is able to solve the key problems in RGB-D fusion and achieves state-of-the-art performance on RGB-D salient object detection.

Laser triggered vacuum switch (LTVS) shows high developing potential compared to many other pulse power switches; To enhance the triggered characteristics of LTVS, this paper adopts different powder mixtures of metal (include Ti, and W) and metal salts (include KCl, NaCl, K 2 CO 3 , and Na 2 CO 3 ) as the target materials of LTVS, while the metal is with favorable thermal conduction capability, and the metal salts is with better chemical activity. The experiment circuits are established later, and the time-delay and triggered jitter of LTVS prototype with different target materials are tested under different laser energy, under-voltage ratio, open distance. And the influences of different target materials on the working life of LTVS are also compared. The study results provide references to the design of LTVS products.

In this paper, we propose a novel calibration framework for the gaze estimation of mobile gaze tracking systems. In our method, the user's eye and the eye camera are modeled as a central catadioptric camera. Thus the epipolar geometry of the mobile gaze tracker can be described by the hybrid two-view geometry. To calibrate this model, the user is asked to gaze at the calibration points distributed in 3-D space but not all located on one plane. In the light of binocular training data, we apply a 3×6 local hybrid-fundamental matrix to register pupil centers with epipolar lines in the scene image. Thus the image gaze point viewed from different depths can be uniquely determined as the intersection of two epipolar lines calculated by binocular data. The simulation and experimental results show the effectiveness of our proposed calibration framework for mobile gaze trackers.