-
Improved far-field speech recognition using Joint Variational Autoencoder
Authors:
Shashi Kumar,
Shakti P. Rath,
Abhishek Pandey
Abstract:
Automatic Speech Recognition (ASR) systems suffer considerably when source speech is corrupted with noise or room impulse responses (RIR). Typically, speech enhancement is applied in both mismatched and matched scenario training and testing. In matched setting, acoustic model (AM) is trained on dereverberated far-field features while in mismatched setting, AM is fixed. In recent past, mapping spee…
▽ More
Automatic Speech Recognition (ASR) systems suffer considerably when source speech is corrupted with noise or room impulse responses (RIR). Typically, speech enhancement is applied in both mismatched and matched scenario training and testing. In matched setting, acoustic model (AM) is trained on dereverberated far-field features while in mismatched setting, AM is fixed. In recent past, mapping speech features from far-field to close-talk using denoising autoencoder (DA) has been explored. In this paper, we focus on matched scenario training and show that the proposed joint VAE based mapping achieves a significant improvement over DA. Specifically, we observe an absolute improvement of 2.5% in word error rate (WER) compared to DA based enhancement and 3.96% compared to AM trained directly on far-field filterbank features.
△ Less
Submitted 24 April, 2022;
originally announced April 2022.
-
A Mixture of Expert Based Deep Neural Network for Improved ASR
Authors:
Vishwanath Pratap Singh,
Shakti P. Rath,
Abhishek Pandey
Abstract:
This paper presents a novel deep learning architecture for acoustic model in the context of Automatic Speech Recognition (ASR), termed as MixNet. Besides the conventional layers, such as fully connected layers in DNN-HMM and memory cells in LSTM-HMM, the model uses two additional layers based on Mixture of Experts (MoE). The first MoE layer operating at the input is based on pre-defined broad phon…
▽ More
This paper presents a novel deep learning architecture for acoustic model in the context of Automatic Speech Recognition (ASR), termed as MixNet. Besides the conventional layers, such as fully connected layers in DNN-HMM and memory cells in LSTM-HMM, the model uses two additional layers based on Mixture of Experts (MoE). The first MoE layer operating at the input is based on pre-defined broad phonetic classes and the second layer operating at the penultimate layer is based on automatically learned acoustic classes. In natural speech, overlap in distribution across different acoustic classes is inevitable, which leads to inter-class mis-classification. The ASR accuracy is expected to improve if the conventional architecture of acoustic model is modified to make them more suitable to account for such overlaps. MixNet is developed keeping this in mind. Analysis conducted by means of scatter diagram verifies that MoE indeed improves the separation between classes that translates to better ASR accuracy. Experiments are conducted on a large vocabulary ASR task which show that the proposed architecture provides 13.6% and 10.0% relative reduction in word error rates compared to the conventional models, namely, DNN and LSTM respectively, trained using sMBR criteria. In comparison to an existing method developed for phone-classification (by Eigen et al), our proposed method yields a significant improvement.
△ Less
Submitted 2 December, 2021;
originally announced December 2021.
-
A higher order Minkowski loss for improved prediction ability of acoustic model in ASR
Authors:
Vishwanath Pratap Singh,
Shakti P. Rath,
Abhishek Pandey
Abstract:
Conventional automatic speech recognition (ASR) system uses second-order minkowski loss during inference time which is suboptimal as it incorporates only first order statistics in posterior estimation [2]. In this paper we have proposed higher order minkowski loss (4th Order and 6th Order) during inference time, without any changes during training time. The main contribution of the paper is to sho…
▽ More
Conventional automatic speech recognition (ASR) system uses second-order minkowski loss during inference time which is suboptimal as it incorporates only first order statistics in posterior estimation [2]. In this paper we have proposed higher order minkowski loss (4th Order and 6th Order) during inference time, without any changes during training time. The main contribution of the paper is to show that higher order loss uses higher order statistics in posterior estimation, which improves the prediction ability of acoustic model in ASR system. We have shown mathematically that posterior probability obtained due to higher order loss is function of second order posterior and thus the method can be incorporated in standard ASR system in an easy manner. It is to be noted that all changes are proposed during test(inference) time, we do not make any change in any training pipeline. Multiple baseline systems namely, TDNN1, TDNN2, DNN and LSTM are developed to verify the improvement incurred due to higher order minkowski loss. All experiments are conducted on LibriSpeech dataset and performance metrics are word error rate (WER) on "dev-clean", "test-clean", "dev-other" and "test-other" datasets.
△ Less
Submitted 2 December, 2021;
originally announced December 2021.
-
Field-theoretical study of the Bose polaron
Authors:
Steffen Patrick Rath,
Richard Schmidt
Abstract:
We study the properties of the Bose polaron, an impurity strongly interacting with a Bose-Einstein condensate, using a field-theoretic approach and make predictions for the spectral function and various quasiparticle properties that can be tested in experiment. We find that most of the spectral weight is contained in a coherent attractive and a metastable repulsive polaron branch. We show that the…
▽ More
We study the properties of the Bose polaron, an impurity strongly interacting with a Bose-Einstein condensate, using a field-theoretic approach and make predictions for the spectral function and various quasiparticle properties that can be tested in experiment. We find that most of the spectral weight is contained in a coherent attractive and a metastable repulsive polaron branch. We show that the qualitative behavior of the Bose polaron is well described by a non-selfconsistent T-matrix approximation by comparing analytical results to numerical data obtained from a fully selfconsistent T-matrix approach. The latter takes into account an infinite number of bosons excited from the condensate.
△ Less
Submitted 16 December, 2013; v1 submitted 15 August, 2013;
originally announced August 2013.
-
Non-local order in Mott insulators, Duality and Wilson Loops
Authors:
Steffen Patrick Rath,
Wolfgang Simeth,
Manuel Endres,
Wilhelm Zwerger
Abstract:
It is shown that the Mott insulating and superfluid phases of bosons in an optical lattice may be distinguished by a non-local 'parity order parameter' which is directly accessible via single site resolution imaging. In one dimension, the lattice Bose model is dual to a classical interface roughening problem. We use known exact results from the latter to prove that the parity order parameter exhib…
▽ More
It is shown that the Mott insulating and superfluid phases of bosons in an optical lattice may be distinguished by a non-local 'parity order parameter' which is directly accessible via single site resolution imaging. In one dimension, the lattice Bose model is dual to a classical interface roughening problem. We use known exact results from the latter to prove that the parity order parameter exhibits long range order in the Mott insulating phase, consistent with recent experiments by Endres et al. [Science 334, 200 (2011)]. In two spatial dimensions, the parity order parameter can be expressed in terms of an equal time Wilson loop of a non-trivial U(1) gauge theory in 2+1 dimensions which exhibits a transition between a Coulomb and a confining phase. The negative logarithm of the parity order parameter obeys a perimeter law in the Mott insulator and is enhanced by a logarithmic factor in the superfluid.
△ Less
Submitted 2 May, 2013; v1 submitted 4 February, 2013;
originally announced February 2013.
-
Efimov physics beyond universality
Authors:
Richard Schmidt,
Steffen Patrick Rath,
Wilhelm Zwerger
Abstract:
We provide an exact solution of the Efimov spectrum in ultracold gases within the standard two-channel model for Feshbach resonances. It is shown that the finite range in the Feshbach coupling makes the introduction of an adjustable three-body parameter obsolete. The solution explains the empirical relation between the scattering length a_- where the first Efimov state appears at the atom threshol…
▽ More
We provide an exact solution of the Efimov spectrum in ultracold gases within the standard two-channel model for Feshbach resonances. It is shown that the finite range in the Feshbach coupling makes the introduction of an adjustable three-body parameter obsolete. The solution explains the empirical relation between the scattering length a_- where the first Efimov state appears at the atom threshold and the van der Waals length l_vdw for open channel dominated resonances. There is a continuous crossover to the closed channel dominated limit, where the scale in the energy level diagram as a function of the inverse scattering length 1/a is set by the intrinsic length r* associated with the Feshbach coupling. Our results provide a number of predictions for non-universal ratios between energies and scattering lengths that can be tested in future experiments.
△ Less
Submitted 17 December, 2012; v1 submitted 20 January, 2012;
originally announced January 2012.
-
Quantum Capillary Waves at the Superfluid--Mott Insulator Interface
Authors:
Steffen Patrick Rath,
Boris Spivak,
Wilhelm Zwerger
Abstract:
We discuss quantum fluctuations of the interface between a superfluid and a Mott-insulating state of ultracold atoms in a trap. The fluctuations of the boundary are due to a new type of surface modes, whose spectrum is similar (but not identical) to classical capillary waves. The corresponding quantum capillary length sets the scale for the penetration of the superfluid into the Mott-insulating re…
▽ More
We discuss quantum fluctuations of the interface between a superfluid and a Mott-insulating state of ultracold atoms in a trap. The fluctuations of the boundary are due to a new type of surface modes, whose spectrum is similar (but not identical) to classical capillary waves. The corresponding quantum capillary length sets the scale for the penetration of the superfluid into the Mott-insulating regime by the proximity effect and may be on the order of several lattice spacings. It determines the typical magnitude of the interface width due to quantum fluctuations, which may be inferred from single site imaging of ultracold atoms in an optical lattice.
△ Less
Submitted 10 October, 2011; v1 submitted 3 August, 2011;
originally announced August 2011.
-
Full counting statistics of the interference contrast from independent Bose-Einstein condensates
Authors:
Steffen Patrick Rath,
Wilhelm Zwerger
Abstract:
We show that the visibility in interference experiments with Bose-Einstein condensates is directly related to the condensate fraction. The probability distribution of the contrast over many runs of an interference experiment thus gives the full counting statistics of the condensed atom number. For two-dimensional Bose gases, we discuss the universal behavior of the probability distribution in the…
▽ More
We show that the visibility in interference experiments with Bose-Einstein condensates is directly related to the condensate fraction. The probability distribution of the contrast over many runs of an interference experiment thus gives the full counting statistics of the condensed atom number. For two-dimensional Bose gases, we discuss the universal behavior of the probability distribution in the superfluid regime and provide analytical expressions for the distributions for both homogeneous and harmonically trapped samples. They are non-Gaussian and unimodal with a variance that is directly related to the superfluid density. In general, the visibility is a self-averaging observable only in the presence of long range phase coherence. Close to the transition temperature, the visibility distribution reflects the universal order parameter distribution in the vicinity of the critical point.
△ Less
Submitted 22 November, 2010; v1 submitted 24 September, 2010;
originally announced September 2010.
-
The equilibrium state of a trapped two-dimensional Bose gas
Authors:
Steffen P. Rath,
Tarik Yefsah,
Kenneth J. Guenter,
Marc Cheneau,
Remi Desbuquois,
Markus Holzmann,
Werner Krauth,
Jean Dalibard
Abstract:
We study experimentally and numerically the equilibrium density profiles of a trapped two-dimensional $^{87}$Rb Bose gas, and investigate the equation of state of the homogeneous system using the local density approximation. We find a clear discrepancy between in-situ measurements and Quantum Monte Carlo simulations, which we attribute to a non-linear variation of the optical density of the atomi…
▽ More
We study experimentally and numerically the equilibrium density profiles of a trapped two-dimensional $^{87}$Rb Bose gas, and investigate the equation of state of the homogeneous system using the local density approximation. We find a clear discrepancy between in-situ measurements and Quantum Monte Carlo simulations, which we attribute to a non-linear variation of the optical density of the atomic cloud with its spatial density. However, good agreement between experiment and theory is recovered for the density profiles measured after time-of-flight, taking advantage of their self-similarity in a two-dimensional expansion.
△ Less
Submitted 23 March, 2010;
originally announced March 2010.
-
Practical scheme for a light-induced gauge field in an atomic Bose gas
Authors:
Kenneth J. Günter,
Marc Cheneau,
Tarik Yefsah,
Steffen P. Rath,
Jean Dalibard
Abstract:
We propose a scheme to generate an Abelian gauge field in an atomic gas using two crossed laser beams. If the internal atomic state follows adiabatically the eigenstates of the atom-laser interaction, Berry's phase gives rise to a vector potential that can nucleate vortices in a Bose gas. The present scheme operates even for a large detuning with respect to the atomic resonance, making it applic…
▽ More
We propose a scheme to generate an Abelian gauge field in an atomic gas using two crossed laser beams. If the internal atomic state follows adiabatically the eigenstates of the atom-laser interaction, Berry's phase gives rise to a vector potential that can nucleate vortices in a Bose gas. The present scheme operates even for a large detuning with respect to the atomic resonance, making it applicable to alkali-metal atoms without significant heating due to spontaneous emission. We test the validity of the adiabatic approximation by integrating the set of coupled Gross-Pitaevskii equations associated with the various internal atomic states, and we show that the steady state of the interacting gas indeed exhibits a vortex lattice, as expected from the adiabatic gauge field.
△ Less
Submitted 3 February, 2009; v1 submitted 24 November, 2008;
originally announced November 2008.
-
Geometric potentials in quantum optics: A semi-classical interpretation
Authors:
Marc Cheneau,
Steffen Patrick Rath,
Tarik Yefsah,
Kenneth John Günter,
Gediminas Juzeliunas,
Jean Dalibard
Abstract:
We propose a semi-classical interpretation of the geometric scalar and vector potentials that arise due to Berry's phase when an atom moves slowly in a light field. Starting from the full quantum Hamiltonian, we turn to a classical description of the atomic centre-of-mass motion while still treating the internal degrees of freedom as quantum variables. We show that the scalar potential can be id…
▽ More
We propose a semi-classical interpretation of the geometric scalar and vector potentials that arise due to Berry's phase when an atom moves slowly in a light field. Starting from the full quantum Hamiltonian, we turn to a classical description of the atomic centre-of-mass motion while still treating the internal degrees of freedom as quantum variables. We show that the scalar potential can be identified as the kinetic energy of an atomic micro-motion caused by quantum fluctuations of the radiative force, and that the Lorentz-type force appears as a result of the motion-induced perturbation of the internal atomic state. For a specific configuration involving two counter-propagating Gaussian laser beams, we relate the geometric forces to the radiation pressure and dipole forces known from quantum optics. The simple physical pictures provided by the present analysis may help for the design and the implementation of novel geometric forces.
△ Less
Submitted 22 August, 2008; v1 submitted 25 July, 2008;
originally announced July 2008.
-
Theory and Applications of Two-dimensional, Null-boundary, Nine-Neighborhood, Cellular Automata Linear rules
Authors:
Pabitra Pal Choudhury,
Birendra Kumar Nayak,
Sudhakar Sahoo,
Sunil Pankaj Rath
Abstract:
This paper deals with the theory and application of 2-Dimensional, nine-neighborhood, null- boundary, uniform as well as hybrid Cellular Automata (2D CA) linear rules in image processing. These rules are classified into nine groups depending upon the number of neighboring cells influences the cell under consideration. All the Uniform rules have been found to be rendering multiple copies of a giv…
▽ More
This paper deals with the theory and application of 2-Dimensional, nine-neighborhood, null- boundary, uniform as well as hybrid Cellular Automata (2D CA) linear rules in image processing. These rules are classified into nine groups depending upon the number of neighboring cells influences the cell under consideration. All the Uniform rules have been found to be rendering multiple copies of a given image depending on the groups to which they belong where as Hybrid rules are also shown to be characterizing the phenomena of zooming in, zooming out, thickening and thinning of a given image. Further, using hybrid CA rules a new searching algorithm is developed called Sweepers algorithm which is found to be applicable to simulate many inter disciplinary research areas like migration of organisms towards a single point destination, Single Attractor and Multiple Attractor Cellular Automata Theory, Pattern Classification and Clustering Problem, Image compression, Encryption and Decryption problems, Density Classification problem etc.
△ Less
Submitted 15 April, 2008;
originally announced April 2008.
-
The trapped two-dimensional Bose gas: from Bose-Einstein condensation to Berezinskii-Kosterlitz-Thouless physics
Authors:
Zoran Hadzibabic,
Peter Krüger,
Marc Cheneau,
Steffen Patrick Rath,
Jean Dalibard
Abstract:
We analyze the results of a recent experiment with bosonic rubidium atoms harmonically confined in a quasi-two-dimensional geometry. In this experiment a well defined critical point was identified, which separates the high-temperature normal state characterized by a single component density distribution, and the low-temperature state characterized by a bimodal density distribution and the emerge…
▽ More
We analyze the results of a recent experiment with bosonic rubidium atoms harmonically confined in a quasi-two-dimensional geometry. In this experiment a well defined critical point was identified, which separates the high-temperature normal state characterized by a single component density distribution, and the low-temperature state characterized by a bimodal density distribution and the emergence of high-contrast interference between independent two-dimensional clouds. We first show that this transition cannot be explained in terms of conventional Bose-Einstein condensation of the trapped ideal Bose gas. Using the local density approximation, we then combine the mean-field (MF) Hartree-Fock theory with the prediction for the Berezinskii-Kosterlitz-Thouless transition in an infinite uniform system. If the gas is treated as a strictly 2D system, the MF predictions for the spatial density profiles significantly deviate from those of a recent Quantum Monte-Carlo (QMC) analysis. However when the residual thermal excitation of the strongly confined degree of freedom is taken into account, an excellent agreement is reached between the MF and the QMC approaches. For the interaction strength corresponding to the experiment, we predict a strong correction to the critical atom number with respect to the ideal gas theory (factor $\sim 2$). A quantitative agreement between theory and experiment is reached concerning the critical atom number if the predicted density profiles are used for temperature calibration.
△ Less
Submitted 25 February, 2008; v1 submitted 8 December, 2007;
originally announced December 2007.
-
Evaporative Cooling of a Guided Rubidium Atomic Beam
Authors:
Thierry Lahaye,
Z. Wang,
G. Reinaudi,
S. P. Rath,
J. Dalibard,
D. Guéry-Odelin
Abstract:
We report on our recent progress in the manipulation and cooling of a magnetically guided, high flux beam of $^{87}{\rm Rb}$ atoms. Typically $7\times 10^9$ atoms per second propagate in a magnetic guide providing a transverse gradient of 800 G/cm, with a temperature $\sim550$ $μ$K, at an initial velocity of 90 cm/s. The atoms are subsequently slowed down to $\sim 60$ cm/s using an upward slope.…
▽ More
We report on our recent progress in the manipulation and cooling of a magnetically guided, high flux beam of $^{87}{\rm Rb}$ atoms. Typically $7\times 10^9$ atoms per second propagate in a magnetic guide providing a transverse gradient of 800 G/cm, with a temperature $\sim550$ $μ$K, at an initial velocity of 90 cm/s. The atoms are subsequently slowed down to $\sim 60$ cm/s using an upward slope. The relatively high collision rate (5 s$^{-1}$) allows us to start forced evaporative cooling of the beam, leading to a reduction of the beam temperature by a factor of ~4, and a ten-fold increase of the on-axis phase-space density.
△ Less
Submitted 30 May, 2005;
originally announced May 2005.