arXiv subjects

Yanzhang He

Publications and source records attributed to Yanzhang He.

51 records · Page 3Linked to original sources

arXiv2019

Two-Pass End-to-End Speech Recognition

The requirements for many applications of state-of-the-art speech recognition systems include not only low word error rate (WER) but also low latency. Specifically, for many use-cases, the system must be able to decode utterances in a streaming fashion and faster than real-time. Recently, a streaming recurrent neural network transducer (RNN-T) end-to-end (E2E) model has shown to be a good candidate for on-device speech recognition, with improved WER and latency metrics compared to conventional on-device models [1]. However, this model still lags behind a large state-of-the-art conventional model in quality [2]. On the other hand, a non-streaming E2E Listen, Attend and Spell (LAS) model has shown comparable quality to large conventional models [3]. This work aims to bring the quality of an E2E streaming model closer to that of a conventional system by incorporating a LAS network as a second-pass component, while still abiding by latency constraints. Our proposed two-pass model achieves a 17%-22% relative reduction in WER compared to RNN-T alone and increases latency by a small fraction over RNN-T.

cs.CL↗

arXiv2019

Lingvo: a Modular and Scalable Framework for Sequence-to-Sequence Modeling

Lingvo is a Tensorflow framework offering a complete solution for collaborative deep learning research, with a particular focus towards sequence-to-sequence models. Lingvo models are composed of modular building blocks that are flexible and easily extensible, and experiment configurations are centralized and highly customizable. Distributed training and quantized inference are supported directly within the framework, and it contains existing implementations of a large number of utilities, helper functions, and the newest research ideas. Lingvo has been used in collaboration by dozens of researchers in more than 20 papers over the last two years. This document outlines the underlying design of Lingvo and serves as an introduction to the various pieces of the framework, while also offering examples of advanced features that showcase the capabilities of the framework.

cs.LG↗

arXiv2018

Streaming End-to-end Speech Recognition For Mobile Devices

End-to-end (E2E) models, which directly predict output character sequences given input speech, are good candidates for on-device speech recognition. E2E models, however, present numerous challenges: In order to be truly useful, such models must decode speech utterances in a streaming fashion, in real time; they must be robust to the long tail of use cases; they must be able to leverage user-specific context (e.g., contact lists); and above all, they must be extremely accurate. In this work, we describe our efforts at building an E2E speech recognizer using a recurrent neural network transducer. In experimental evaluations, we find that the proposed approach can outperform a conventional CTC-based model in terms of both latency and accuracy in a number of evaluation categories.

cs.CL↗

arXiv2017

Streaming Small-Footprint Keyword Spotting using Sequence-to-Sequence Models

We develop streaming keyword spotting systems using a recurrent neural network transducer (RNN-T) model: an all-neural, end-to-end trained, sequence-to-sequence model which jointly learns acoustic and language model components. Our models are trained to predict either phonemes or graphemes as subword units, thus allowing us to detect arbitrary keyword phrases, without any out-of-vocabulary words. In order to adapt the models to the requirements of keyword spotting, we propose a novel technique which biases the RNN-T system towards a specific keyword of interest. Our systems are compared against a strong sequence-trained, connectionist temporal classification (CTC) based "keyword-filler" baseline, which is augmented with a separate phoneme language model. Overall, our RNN-T system with the proposed biasing technique significantly improves performance over the baseline system.

cs.CL↗

arXiv2013

A 2-level system of cold sodium atoms with high and tunable susceptibility against the magnetic field

A 2-level spin-system of cold sodium atoms is proposed. Both the cases that the system has arrived at thermo-equilibrium and is in the early stage of evolution have been studied. This system is inert to the magnetic field $B$ in general but very sensitive in a narrow domain around $B=B_0$, where $B_0$ can be predicted and is tunable. A characteristic constant $γ=0.278466$ dedicated to various 2-level systems is found, and leads to a upper limit for the internal energy $U$ of the whole system so that $U\leqγk_B T$. This limit is considerably lower than the energy assigned to the spatial motion of only a single particle. Under thermo-equilibrium, the populations of spin-components measured at distinct $T$ converge to a fixed value when $B=B_0$. The period and amplitude of population oscillation are found to be seriously affected by the special sensitivity against $B$. Rich messages on the dynamic parameters could be obtained via experimental measurements of these systems.

physics.atom-ph↗

arXiv2012

Universality in tightly bound 3-boson systems

The effects of two distinct operations of the elements of the symmetry groups of a Hamiltonian on a quantum state might be equivalent in some specific zones of coordinate space. Making use of the matrix representations of the groups, the equivalence leads to a set of homogeneous linear equations imposing on the wave functions. When the matrix of the equations is non-degenerate, the wave functions will appear as nodal surfaces in these zones. Therefore, the equivalence leads to the existence of inherent nodal structure in the quantum states. In this paper, trapped 3-boson systems with different types of interactions are studied. The structures of the tightly bound eigenstates have been analyzed systematically. The emphasis is placed to demonstrate the universality arising from the common inherent nodal structures.

quant-ph↗

arXiv2011

Elementary modes of excitation caused by the quadratic Zeeman term and the sensitivity of spin structures of small spin-2 condensates against the magnetic field

The response of spin-2 small condensates to an external magnetic field $B$ is studied. The parameters of the interaction are considered as variable. The emphasis is placed on clarifying the modes of excitation caused by the quadratic Zeeman term. The theoretical method used is beyond the mean field theory. A set of eigenstates with the $U(5)\supset SO(5)\supset SO(3)$ symmetry is introduced to facilitate the analysis. To obtain a quantitative evaluation on the response, the fidelity susceptibility and the $B$ -dependent average populations of spin-components have been calculated. Mostly the particle number N=30 is assumed. The effect with a larger or smaller $N$ is also considered. It was found that the sensitivity of the response depends strongly both on the interaction and on the inherent symmetry.

cond-mat.quant-gas↗

arXiv2010

Repeating head-on collisions in an optical trap and the evaluation of spin-dependent interactions among neutral particles

A dynamic process of repeating collisions of a pair of trapped neutral particles with weak spin-dependent interaction is designed and studied. Related theoretical derivation and numerical calculation have been performed to study the inherent coordinate-spin and momentum-spin correlation. Due to the repeating collisions the effect of the weak interaction can be accumulated and enlarged, and therefore can be eventually detected. Numerical results suggest that the Cr-Cr interaction, which has not yet been completely clear, could be thereby determined. The design can be in general used to determine various interactions among neutral atoms and molecules, in particular for the determination of very weak forces.

cond-mat.other↗

arXiv2010

Two-body scattering in a trap and a special periodic phenomenon sensitive to the interaction

Two-body scattering of neutral particles in a trap is studied theoretically. The control of the initial state is realized by using optical traps. The collisions inside the trap occur repeatedly; thereby the effect of interaction can be accumulated. Two periodic phenomena with a shorter and a much longer period, respectively, are found. The latter is sensitive to the interaction. Instead of measuring the differential cross section as usually does, the measurement of the longer period and the details of the periodic behavior might be a valid source of information on weak interactions among neutral particles.

cond-mat.other↗

arXiv2010

Periodic giant-persistent current in sharp pulses on a ring

We show here a mesoscopic device based on a narrow ring containing an electron. In the device, an amount of energy is stored in advance. Similar to the pendulum, an exact periodic motion of the electron is thereby initiated afterward. The motion appears as a series of sharp pulses, and is in nature different from the well known Aharonov-Bohm (A-B) oscillation. In particular, the pulses of current can be tuned to be very strong (say, more than two orders stronger than the usual A-B current). Related theory and numerical results are presented.

cond-mat.other↗

arXiv2009

The spin evolution of spin-3 $^{52}$Cr Bose-Einstein condensate

The spin evolution of a Bose-Einstein condensate starting from a mixture of two or three groups of $^{52}$Cr (spin-3) atoms in an optical trap has been studied theoretically. The initial state is so chosen that the system does not distinguish up and down. In this choice, the deviation caused by the single-mode approximation is reduced. Moreover, since the particle number is given very small (N=20), the deviation caused by the neglect of the long-range dipole force is also reduced. Making use of these two simplifications, a theoretical calculation beyond the mean field theory is performed. The numerical results are help to evaluate the unknown strength $g_0$.

cond-mat.quant-gas↗

arXiv2009

Strong spin-oscillation of small spin-1 condensates caused by an inclined weak magnetic field

When a magnetic field is applied along a direction deviated from the quantization $Z$-axis, the conservation of total magnetization holds no more. In this case the inclined field can cause a strong spin-evolution via the linear Zeeman term even the field is as weak as a percentage of $mG$. An approach beyond the mean field theory is proposed to study the evolution of small $^{87}$Rb condensates under the weak inclined fields. The time-dependent populations of spin-components are given in analytical forms. The evolution is found to be highly sensitive to the magnitude and direction of the field.

cond-mat.other↗

arXiv2008

Mesoscopic oscillator in U-shape with giant persistent current

A mesoscopic oscillator in U-shape has been proposed and studied. Making use of a magnetic flux together with a potential of confinement, the electron contained in the oscillator has been localized initially and an amount of energy has been thereby stored. Then a sudden cancellation of both the potential and the flux may cause an initial current which initiates a periodic motion of the electron from one end of the U-oscillator to the opposite end, and repeatedly. The period is adjustable. The current associated with the periodic motion can be tuned very strong (say, more than two orders larger than the current of the usual Aharonov-Bohm oscillation). Related theory and numerical results are presented.

cond-mat.mes-hall↗

arXiv2008

Fractional Aharonov-Bohm oscillation of a two-layer ring with two electrons

When a circular ring suffers a special topological transformation, it becomes a two-layer ring. Due to the special topology of the two-layer ring, orbital angular momenta are allowed to be a half-integer, this would affect the traditional Aharonov-Bohm oscillation (ABO). In this paper the fractional ABO of the ground state energy, persistent current, and dipole transition of a two-layer ring with two electrons has been studied. Collective and internal coordinates $θ_{C}$ and $ϕ$ have been introduced. Based on them a very simple formula for the current has been obtained, the symmetry constraint imposed on the dipole transition has been clarified, a strict relation between the photon energies of the dipole radiation and the persistent current of the ground state has been found. Comparing with the one-layer rings, the period of the fractional ABO of the two-layer rings becomes much shorter.

cond-mat.other↗

arXiv2007

Density oscillation in highly flattened quantum elliptic rings and tunable strong dipole radiation

A narrow elliptic ring containing an electron threaded by a magnetic field B is studied. When the ring is highly flattened, the increase of B would lead to a big energy gap between the ground and excited states, and therefore lead to a strong emission of dipole photons. The photon frequency can be tuned in a wide range by changing B and/or the shape of the ellipse. The particle density is found to oscillate from a pattern of distribution to another pattern back and forth against $B$. This is a new kind of Aharonov-Bohm oscillation originating from symmetry breaking and is different from the usual oscillation of persistent current.

cond-mat.other↗