arXiv subjects

Zhuo Chen

Publications and source records attributed to Zhuo Chen.

At least 415 records · Page 23Linked to original sources

arXiv2019

Unsupervised Adaptation with Domain Separation Networks for Robust Speech Recognition

Unsupervised domain adaptation of speech signal aims at adapting a well-trained source-domain acoustic model to the unlabeled data from target domain. This can be achieved by adversarial training of deep neural network (DNN) acoustic models to learn an intermediate deep representation that is both senone-discriminative and domain-invariant. Specifically, the DNN is trained to jointly optimize the primary task of senone classification and the secondary task of domain classification with adversarial objective functions. In this work, instead of only focusing on learning a domain-invariant feature (i.e. the shared component between domains), we also characterize the difference between the source and target domain distributions by explicitly modeling the private component of each domain through a private component extractor DNN. The private component is trained to be orthogonal with the shared component and thus implicitly increases the degree of domain-invariance of the shared component. A reconstructor DNN is used to reconstruct the original speech feature from the private and shared components as a regularization. This domain separation framework is applied to the unsupervised environment adaptation task and achieved 11.08% relative WER reduction from the gradient reversal layer training, a representative adversarial training method, for automatic speech recognition on CHiME-3 dataset.

cs.CL↗

arXiv2019

Speaker-Invariant Training via Adversarial Learning

We propose a novel adversarial multi-task learning scheme, aiming at actively curtailing the inter-talker feature variability while maximizing its senone discriminability so as to enhance the performance of a deep neural network (DNN) based ASR system. We call the scheme speaker-invariant training (SIT). In SIT, a DNN acoustic model and a speaker classifier network are jointly optimized to minimize the senone (tied triphone state) classification loss, and simultaneously mini-maximize the speaker classification loss. A speaker-invariant and senone-discriminative deep feature is learned through this adversarial multi-task learning. With SIT, a canonical DNN acoustic model with significantly reduced variance in its output probabilities is learned with no explicit speaker-independent (SI) transformations or speaker-specific representations used in training or testing. Evaluated on the CHiME-3 dataset, the SIT achieves 4.99% relative word error rate (WER) improvement over the conventional SI acoustic model. With additional unsupervised speaker adaptation, the speaker-adapted (SA) SIT model achieves 4.86% relative WER gain over the SA SI acoustic model.

eess.AS↗

arXiv2019

Low-Latency Speaker-Independent Continuous Speech Separation

Speaker independent continuous speech separation (SI-CSS) is a task of converting a continuous audio stream, which may contain overlapping voices of unknown speakers, into a fixed number of continuous signals each of which contains no overlapping speech segment. A separated, or cleaned, version of each utterance is generated from one of SI-CSS's output channels nondeterministically without being split up and distributed to multiple channels. A typical application scenario is transcribing multi-party conversations, such as meetings, recorded with microphone arrays. The output signals can be simply sent to a speech recognition engine because they do not include speech overlaps. The previous SI-CSS method uses a neural network trained with permutation invariant training and a data-driven beamformer and thus requires much processing latency. This paper proposes a low-latency SI-CSS method whose performance is comparable to that of the previous method in a microphone array-based meeting transcription task.This is achieved (1) by using a new speech separation network architecture combined with a double buffering scheme and (2) by performing enhancement with a set of fixed beamformers followed by a neural post-filter.

eess.AS↗

arXiv2019

Envisioning the next decade of Galactic Center science: a laboratory for the study of the physics and astrophysics of supermassive black holes

As the closest example of a galactic nucleus, the Galactic center (GC) presents an exquisite laboratory for learning about supermassive black holes (SMBH) and their environment. We describe several exciting new research directions that, over the next 10 years, hold the potential to answer some of the biggest scientific questions raised in recent decades: Is General Relativity (GR) the correct description for supermassive black holes? What is the nature of star formation in extreme environments? How do stars and compact objects dynamically interact with the supermassive black hole? What physical processes drive gas accretion in low-luminosity black holes? We describe how the high sensitivity, angular resolution, and astrometric precision offered by the next generation of large ground-based telescopes with adaptive optics will help us answer these questions. First, it will be possible to obtain precision measurements of stellar orbits in the Galaxy's central potential, providing both tests of GR in the unexplored regime near a SMBH and measurements of the extended dark matter distribution that is predicted to exist at the GC. Second, we will probe stellar populations at the GC to significantly lower masses than are possible today, down to brown dwarfs. Their structure and dynamics will provide an unprecedented view of the stellar cusp around the SMBH and will distinguish between models of star formation in this extreme environment. This increase in depth will also allow us to measure the currently unknown population of compact remnants at the GC by observing their effects on luminous sources. Third, uncertainties on the mass of and distance to the SMBH can be improved by a factor of $\sim$10. Finally, we can also study the near-infrared accretion onto the black hole at unprecedented sensitivity and time resolution, which can reveal the underlying physics of black hole accretion.

astro-ph.GA↗

arXiv2019

On the Distribution of GSVD

In this paper, some new results on the distribution of the generalized singular value decomposition (GSVD) are presented.

cs.IT↗

arXiv2019

Understanding the Impact of Label Granularity on CNN-based Image Classification

In recent years, supervised learning using Convolutional Neural Networks (CNNs) has achieved great success in image classification tasks, and large scale labeled datasets have contributed significantly to this achievement. However, the definition of a label is often application dependent. For example, an image of a cat can be labeled as "cat" or perhaps more specifically "Persian cat." We refer to this as label granularity. In this paper, we conduct extensive experiments using various datasets to demonstrate and analyze how and why training based on fine-grain labeling, such as "Persian cat" can improve CNN accuracy on classifying coarse-grain classes, in this case "cat." The experimental results show that training CNNs with fine-grain labels improves both network's optimization and generalization capabilities, as intuitively it encourages the network to learn more features, and hence increases classification accuracy on coarse-grain classes under all datasets considered. Moreover, fine-grain labels enhance data efficiency in CNN training. For example, a CNN trained with fine-grain labels and only 40% of the total training data can achieve higher accuracy than a CNN trained with the full training dataset and coarse-grain labels. These results point to two possible applications of this work: (i) with sufficient human resources, one can improve CNN performance by re-labeling the dataset with fine-grain labels, and (ii) with limited human resources, to improve CNN performance, rather than collecting more training data, one may instead use fine-grain labels for the dataset. We further propose a metric called Average Confusion Ratio to characterize the effectiveness of fine-grain labeling, and show its use through extensive experimentation. Code is available at https://github.com/cmu-enyac/Label-Granularity.

cs.CV↗

arXiv2018

Shifted derived Poisson manifolds associated with Lie pairs

We study the shifted analogue of the "Lie--Poisson" construction for $L_\infty$ algebroids and we prove that any $L_\infty$ algebroid naturally gives rise to shifted derived Poisson manifolds. We also investigate derived Poisson structures from a purely algebraic perspective and, in particular, we establish a homotopy transfer theorem for derived Poisson algebras. As an application, we prove that, given a Lie pair $(L,A)$, the space $\operatorname{tot}Ω^{\bullet}_A(Λ^\bullet(L/A))$ admits a degree $(+1)$ derived Poisson algebra structure with the wedge product as associative multiplication and the Chevalley--Eilenberg differential $d_A^{\operatorname{Bott}}:Ω^{\bullet}_A(Λ^\bullet(L/A))\to Ω^{\bullet +1}_A(Λ^\bullet(L/A))$ as unary $L_\infty$ bracket. This degree $(+1)$ derived Poisson algebra structure on $\operatorname{tot}Ω^{\bullet}_A(Λ^\bullet(L/A))$ is unique up to an isomorphism having the identity map as first Taylor coefficient. Consequently, the Chevalley--Eilenberg hypercohomology $\mathbb{H}(Ω^{\bullet}_A(Λ^\bullet(L/A)),d_A^{\operatorname{Bott}})$ admits a canonical Gerstenhaber algebra structure.

math.QA↗

arXiv2018

Mobile Collaborative Spectrum Sensing for Heterogeneous Networks: A Bayesian Machine Learning Approach

Spectrum sensing in a large-scale heterogeneous network is very challenging as it usually requires a large number of static secondary users (SUs) to obtain the global spectrum states. To tackle this problem, in this paper, we propose a new framework based on Bayesian machine learning. We exploit the mobility of multiple SUs to simultaneously collect spectrum sensing data, and cooperatively derive the global spectrum states. We first develop a novel non-parametric Bayesian learning model, referred to as beta process sticky hidden Markov model (BP-SHMM), to capture the spatial-temporal correlation in the collected spectrum data, where SHMM models the latent statistical correlation within each mobile SU's time series data, while BP realizes the cooperation among multiple SUs. Bayesian inference is then carried out to automatically infer the heterogeneous spectrum states. Based on the inference results, we also develop a new algorithm with a refinement mechanism to predict the spectrum availability, which enables a newly joining SU to immediately access the unoccupied frequency band without sensing. Simulation results show that the proposed framework can significantly improve spectrum sensing performance compared with the existing spectrum sensing techniques.

cs.IT↗

arXiv2018

Accretion in Common Envelope Evolution

Common envelope evolution (CEE) occurs in some binary systems involving asymptotic giant branch (AGB) or red giant branch (RGB) stars, and understanding this process is crucial for understanding the origins of various transient phenomena. CEE has been shown to be highly asymmetrical and global 3D simulations are needed to help understand the dynamics. We perform and analyze hydrodynamic CEE simulations with the adaptive mesh refinement (AMR) code AstroBEAR, and focus on the role of accretion onto the companion star. We bracket the range of accretion rates by comparing a model that removes mass and pressure using a subgrid accretion prescription with one that does not. Provided a pressure-release valve, such as a bipolar jet, is available, super-Eddington accretion could be common. Finally, we summarize new results pertaining to the energy budget, and discuss the overall implications relating to the feasibility of unbinding the envelope in CEE simulations.

astro-ph.SR↗

arXiv2018

Recognizing Overlapped Speech in Meetings: A Multichannel Separation Approach Using Neural Networks

The goal of this work is to develop a meeting transcription system that can recognize speech even when utterances of different speakers are overlapped. While speech overlaps have been regarded as a major obstacle in accurately transcribing meetings, a traditional beamformer with a single output has been exclusively used because previously proposed speech separation techniques have critical constraints for application to real meetings. This paper proposes a new signal processing module, called an unmixing transducer, and describes its implementation using a windowed BLSTM. The unmixing transducer has a fixed number, say J, of output channels, where J may be different from the number of meeting attendees, and transforms an input multi-channel acoustic signal into J time-synchronous audio streams. Each utterance in the meeting is separated and emitted from one of the output channels. Then, each output signal can be simply fed to a speech recognition back-end for segmentation and transcription. Our meeting transcription system using the unmixing transducer outperforms a system based on a state-of-the-art neural mask-based beamformer by 10.8%. Significant improvements are observed in overlapped segments. To the best of our knowledge, this is the first report that applies overlapped speech recognition to unconstrained real meeting audio.

eess.AS↗

arXiv2018

Intermediate Deep Feature Compression: the Next Battlefield of Intelligent Sensing

The recent advances of hardware technology have made the intelligent analysis equipped at the front-end with deep learning more prevailing and practical. To better enable the intelligent sensing at the front-end, instead of compressing and transmitting visual signals or the ultimately utilized top-layer deep learning features, we propose to compactly represent and convey the intermediate-layer deep learning features of high generalization capability, to facilitate the collaborating approach between front and cloud ends. This strategy enables a good balance among the computational load, transmission load and the generalization ability for cloud servers when deploying the deep neural networks for large scale cloud based visual analysis. Moreover, the presented strategy also makes the standardization of deep feature coding more feasible and promising, as a series of tasks can simultaneously benefit from the transmitted intermediate layers. We also present the results for evaluation of lossless deep feature compression with four benchmark data compression methods, which provides meaningful investigations and baselines for future research and standardization activities.

cs.MM↗

arXiv2018

Optical Lattices with Higher-order Exceptional Points by Non-Hermitian Coupling

Exceptional points (EPs) are degeneracies in open wave systems with coalescence of at least two energy levels and their corresponding eigenstates. In higher dimensions, more complex EP physics not found in two-state systems is observed. We consider the emergence and interaction of multiple EPs in a four coupled optical waveguides system by non-Hermitian coupling showing a unique EP formation pattern in a phase diagram. In addition, absolute phase rigidities are computed to show the mixing of the different states in definite parameter regimes. Our results could be potentially important for developing further understanding of EP physics in higher dimensions via generalized paradigm of nonHermitian coupling for a new generation of parity-time (PT) devices.

physics.optics↗

arXiv2018

Planetary Nebulae Shaped By Common Envelope Evolution

The morphologies of planetary nebula have long been believed to be due to wind shaping processes in which a fast wind from the central star impacts a previously ejected envelope. Asymmetries assumed to exist in the slow wind envelope lead to inertial confinement shaping the resulting interacting winds flow. We present new results demonstrating the effectiveness of Common Envelope Evolution (CEE) at producing aspherical envelopes which, when impinged upon by a spherical fast stellar wind, produce highly bipolar, jet-like outflows. We have run two simple cases using the output of a single PHANTOM SPH CEE simulation. Our work uses the Adaptive Mesh Refinement code AstroBEAR to track the interaction of the fast wind and CEE ejecta allowing us to follow the morphological evolution of the outflow lobes at high resolution in 3-D. Our two models bracket low and high momentum output fast winds. We find the interaction leads to highly collimated bipolar outflows. In addition, the bipolar morphology depends on the fast wind momentum injection rate. With this dependence comes the initiation of significant symmetry breaking between the top and bottom bipolar lobes. Our simulations, though simplified, confirm the long-standing belief that CEE can plan a major role in PPN and PN shaping.

astro-ph.SR↗

arXiv2018

Landau-Ginzburg models for certain fiber products with curves

In this paper we describe a physical realization of a family of non-compact Kahler threefolds with trivial canonical bundle in hybrid Landau-Ginzburg models, motivated by some recent non-Kahler solutions of Strominger systems, and utilizing some recent ideas from GLSMs. We consider threefolds given as fiber products of compact genus g Riemann surfaces and noncompact threefolds. Each genus g Riemann surface is constructed using recent GLSM tricks, as a double cover of P^1 branched over a degree 2g + 2 locus, realized via nonperturbative effects rather than as the critical locus of a superpotential. We focus in particular on special cases corresponding to a set of Kahler twistor spaces of certain hyperKahler four-manifolds, specifically the twistor spaces of R^4, C^2/Z_k, and S^1 x R^3. We check in all cases that the condition for trivial canonical bundle arising physically matches the mathematical constraint.

hep-th↗

arXiv2018

Asymptotic Performance Analysis of GSVD-NOMA Systems with a Large-Scale Antenna Array

This paper considers a multiple-input multiple-output (MIMO) downlink communication scenario with one base station and two users, where each user is equipped with m antennas and the base station is equipped with n antennas. To efficiently exploit the spectrum resources, we propose a transmission protocol which combines generalized singular value decomposition (GSVD) and non-orthogonal multiple access (NOMA). The average data rates achieved by the two users are adopted as performance metrics for evaluation of the proposed GSVD-NOMA scheme. In particular, we first characterize the limiting distribution of the squared generalized singular values of the two users' channel matrices for the asymptotic case where the numbers of transmit and receive antennas approach infinity. Then, we calculate the normalized average individual rates of the users in the considered asymptotic regime. Furthermore, we extend the proposed GSVD-NOMA scheme to the MIMO downlink communication scenario with more than two users by using a hybrid multiple access (MA) approach, where the base station first divides the users into different groups, then the proposed GSVD-NOMA scheme is implemented within each group, and different groups are allocated with orthogonal bandwidth resources. Finally, numerical results are provided to validate the effectiveness of the proposed GSVD-NOMA protocol, and the accuracy of the developed analytical results.

cs.IT↗

arXiv2018

Accretion in common envelope evolution

Common envelope evolution (CEE) is presently a poorly understood, yet critical, process in binary stellar evolution. Characterizing the full 3D dynamics of CEE is difficult in part because simulating CEE is so computationally demanding. Numerical studies have yet to conclusively determine how the envelope ejects and a tight binary results, if only the binary potential energy is used to propel the envelope. Additional power sources might be necessary and accretion onto the inspiraling companion is one such source. Accretion is likely common in post-asymptotic giant branch (AGB) binary interactions but how it operates and how its consequences depend on binary separation remain open questions. Here we use high resolution global 3D hydrodynamic simulations of CEE with the adaptive mesh refinement (AMR) code AstroBEAR, to bracket the range of CEE companion accretion rates by comparing runs that remove mass and pressure via a subgrid accretion model with those that do not. The results show that if a pressure release valve is available, super-Eddington accretion may be common. Jets are a plausible release valve in these environments, and they could also help unbind and shape the envelopes.

astro-ph.SR↗

arXiv2018

Speaker-independent Speech Separation with Deep Attractor Network

Despite the recent success of deep learning for many speech processing tasks, single-microphone, speaker-independent speech separation remains challenging for two main reasons. The first reason is the arbitrary order of the target and masker speakers in the mixture permutation problem, and the second is the unknown number of speakers in the mixture output dimension problem. We propose a novel deep learning framework for speech separation that addresses both of these issues. We use a neural network to project the time-frequency representation of the mixture signal into a high-dimensional embedding space. A reference point attractor is created in the embedding space to represent each speaker which is defined as the centroid of the speaker in the embedding space. The time-frequency embeddings of each speaker are then forced to cluster around the corresponding attractor point which is used to determine the time-frequency assignment of the speaker. We propose three methods for finding the attractors for each source in the embedding space and compare their advantages and limitations. The objective function for the network is standard signal reconstruction error which enables end-to-end operation during both training and test phases. We evaluated our system using the Wall Street Journal dataset WSJ0 on two and three speaker mixtures and report comparable or better performance than other state-of-the-art deep learning methods for speech separation.

cs.SD↗

arXiv2018

Developing Far-Field Speaker System Via Teacher-Student Learning

In this study, we develop the keyword spotting (KWS) and acoustic model (AM) components in a far-field speaker system. Specifically, we use teacher-student (T/S) learning to adapt a close-talk well-trained production AM to far-field by using parallel close-talk and simulated far-field data. We also use T/S learning to compress a large-size KWS model into a small-size one to fit the device computational cost. Without the need of transcription, T/S learning well utilizes untranscribed data to boost the model performance in both the AM adaptation and KWS model compression. We further optimize the models with sequence discriminative training and live data to reach the best performance of systems. The adapted AM improved from the baseline by 72.60% and 57.16% relative word error rate reduction on play-back and live test data, respectively. The final KWS model size was reduced by 27 times from a large-size KWS model without losing accuracy.

cs.CL↗