-
A Comprehensive Solution to Connect Speech Encoder and Large Language Model for ASR
Authors:
Van Tung Pham,
Yist Lin,
Tao Han,
Wei Li,
Jun Zhang,
Lu Lu,
Yuxuan Wang
Abstract:
Recent works have shown promising results in connecting speech encoders to large language models (LLMs) for speech recognition. However, several limitations persist, including limited fine-tuning options, a lack of mechanisms to enforce speech-text alignment, and high insertion errors especially in domain mismatch conditions. This paper presents a comprehensive solution to address these issues. We…
▽ More
Recent works have shown promising results in connecting speech encoders to large language models (LLMs) for speech recognition. However, several limitations persist, including limited fine-tuning options, a lack of mechanisms to enforce speech-text alignment, and high insertion errors especially in domain mismatch conditions. This paper presents a comprehensive solution to address these issues. We begin by investigating more thoughtful fine-tuning schemes. Next, we propose a matching loss to enhance alignment between modalities. Finally, we explore training and inference methods to mitigate high insertion errors. Experimental results on the Librispeech corpus demonstrate that partially fine-tuning the encoder and LLM using parameter-efficient methods, such as LoRA, is the most cost-effective approach. Additionally, the matching loss improves modality alignment, enhancing performance. The proposed training and inference methods significantly reduce insertion errors.
△ Less
Submitted 25 June, 2024;
originally announced June 2024.
-
Fast current-induced skyrmion motion in synthetic antiferromagnets
Authors:
Van Tuong Pham,
Naveen Sisodia,
Ilaria Di Manici,
Joseba Urrestarazu-Larrañaga,
Kaushik Bairagi,
Johan Pelloux-Prayer,
Rodrigo Guedas,
Liliana Buda-Prejbeanu,
Stéphane Auffret,
Andrea Locatelli,
Tevfik Onur Menteş,
Stefania Pizzini,
Pawan Kumar,
Aurore Finco,
Vincent Jacques,
Gilles Gaudin,
Olivier Boulle
Abstract:
Magnetic skyrmions are topological magnetic textures that hold great promise as nanoscale bits of information in memory and logic devices. Although room-temperature ferromagnetic skyrmions and their current-induced manipulation have been demonstrated, their velocity has been limited to about 100 meters per second. In addition, their dynamics are perturbed by the skyrmion Hall effect, a motion tran…
▽ More
Magnetic skyrmions are topological magnetic textures that hold great promise as nanoscale bits of information in memory and logic devices. Although room-temperature ferromagnetic skyrmions and their current-induced manipulation have been demonstrated, their velocity has been limited to about 100 meters per second. In addition, their dynamics are perturbed by the skyrmion Hall effect, a motion transverse to the current direction caused by the skyrmion topological charge. Here, we show that skyrmions in compensated synthetic antiferromagnets can be moved by current along the current direction at velocities of up to 900 meters per second. This can be explained by the cancellation of the net topological charge leading to a vanishing skyrmion Hall effect. Our results open an important path toward the realization of logic and memory devices based on the fast manipulation of skyrmions in tracks.
△ Less
Submitted 19 April, 2024;
originally announced April 2024.
-
Joint Alignment of Multivariate Quasi-Periodic Functional Data Using Deep Learning
Authors:
Vi Thanh Pham,
Jonas Bille Nielsen,
Klaus Fuglsang Kofoed,
Jørgen Tobias Kühl,
Andreas Kryger Jensen
Abstract:
The joint alignment of multivariate functional data plays an important role in various fields such as signal processing, neuroscience and medicine, including the statistical analysis of data from wearable devices. Traditional methods often ignore the phase variability and instead focus on the variability in the observed amplitude. We present a novel method for joint alignment of multivariate quasi…
▽ More
The joint alignment of multivariate functional data plays an important role in various fields such as signal processing, neuroscience and medicine, including the statistical analysis of data from wearable devices. Traditional methods often ignore the phase variability and instead focus on the variability in the observed amplitude. We present a novel method for joint alignment of multivariate quasi-periodic functions using deep neural networks, decomposing, but retaining all the information in the data by preserving both phase and amplitude variability. Our proposed neural network uses a special activation of the output that builds on the unit simplex transformation, and we utilize a loss function based on the Fisher-Rao metric to train our model. Furthermore, our method is unsupervised and can provide an optimal common template function as well as subject-specific templates. We demonstrate our method on two simulated datasets and one real example, comprising data from 12-lead 10s electrocardiogram recordings.
△ Less
Submitted 14 November, 2023;
originally announced December 2023.
-
RdimKD: Generic Distillation Paradigm by Dimensionality Reduction
Authors:
Yi Guo,
Yiqian He,
Xiaoyang Li,
Haotong Qin,
Van Tung Pham,
Yang Zhang,
Shouda Liu
Abstract:
Knowledge Distillation (KD) emerges as one of the most promising compression technologies to run advanced deep neural networks on resource-limited devices. In order to train a small network (student) under the guidance of a large network (teacher), the intuitive method is regularizing the feature maps or logits of the student using the teacher's information. However, existing methods either over-r…
▽ More
Knowledge Distillation (KD) emerges as one of the most promising compression technologies to run advanced deep neural networks on resource-limited devices. In order to train a small network (student) under the guidance of a large network (teacher), the intuitive method is regularizing the feature maps or logits of the student using the teacher's information. However, existing methods either over-restrict the student to learn all information from the teacher, which lead to some bad local minimum, or use various fancy and elaborate modules to process and align features, which are complex and lack generality. In this work, we proposed an abstract and general paradigm for the KD task, referred to as DIMensionality Reduction KD (RdimKD), which solely relies on dimensionality reduction, with a very minor modification to naive L2 loss. RdimKD straightforwardly utilizes a projection matrix to project both the teacher's and student's feature maps onto a low-dimensional subspace, which are then optimized during training. RdimKD achieves the goal in the simplest way that not only does the student get valuable information from the teacher, but it also ensures sufficient flexibility to adapt to the student's low-capacity reality. Our extensive empirical findings indicate the effectiveness of RdimKD across various learning tasks and diverse network architectures.
△ Less
Submitted 14 December, 2023;
originally announced December 2023.
-
Quantification of spin-charge interconversion in highly resistive sputtered Bi$_x$Se$_{1-x}$ with non-local spin valves
Authors:
Isabel C. Arango,
Won Young Choi,
Van Tuong Pham,
Inge Groen,
Diogo C. Vaz,
Punyashloka Debashis,
Hai Li,
Mahendra DC,
Kaan Oguz,
Andrey Chuvilin,
Luis E. Hueso,
Ian A. Young,
Fèlix Casanova
Abstract:
The development of spin-orbitronic devices, such as magneto-electric spin-orbit logic devices, calls for materials with a high resistivity and a high spin-charge interconversion efficiency. One of the most promising candidates in this regard is sputtered Bi$_x$Se$_{1-x}$. Although there are several techniques to quantify spin-charge interconversion, to date reported values for sputtered Bi$_x$Se…
▽ More
The development of spin-orbitronic devices, such as magneto-electric spin-orbit logic devices, calls for materials with a high resistivity and a high spin-charge interconversion efficiency. One of the most promising candidates in this regard is sputtered Bi$_x$Se$_{1-x}$. Although there are several techniques to quantify spin-charge interconversion, to date reported values for sputtered Bi$_x$Se$_{1-x}$ have often been overestimated due to spurious effects related to local currents combined with a lack of understanding of the effect of the interfaces and the use of approximations for unknown parameters, such as the spin diffusion length. In the present study, non-local spin valves are used to inject pure spin currents into Bi$_x$Se$_{1-x}$, allowing us to directly obtain its spin diffusion length as well as its spin Hall angle, from 10 K up to 300 K. These values, which are more accurate than those previously reported in sputtered Bi$_x$Se$_{1-x}$, evidence that the efficiency of this material is not exceptional. Indeed, the figure of merit for spin-charge interconversion, given by the product of these two parameters, is slightly under 1 nm. Our work demonstrates the importance of considering all material parameters and interfaces when quantifying the spin transport properties of materials with strong spin-orbit coupling.
△ Less
Submitted 6 November, 2023;
originally announced November 2023.
-
Electrical detection and nucleation of a magnetic skyrmion in a magnetic tunnel junction observed via operando magnetic microscopy
Authors:
by J. Urrestarazu Larrañaga,
Naveen Sisodia,
Van Tuong Pham,
Ilaria Di Manici,
Aurélien Masseboeuf,
Kevin Garello,
Florian Disdier,
Bruno Fernandez,
Sebastian Wintz,
Markus Weigand,
Mohamed Belmeguenai,
Stefania Pizzini,
Ricardo Sousa,
Liliana Buda-Prejbeanu,
Gilles Gaudin,
Olivier Boulle
Abstract:
Magnetic skyrmions are topological spin textures which are envisioned as nanometre scale information carriers in magnetic memory and logic devices. The recent demonstration of room temperature stabilization of skyrmions and their current induced manipulation in industry compatible ultrathin films were first steps towards the realisation of such devices. However, important challenges remain regardi…
▽ More
Magnetic skyrmions are topological spin textures which are envisioned as nanometre scale information carriers in magnetic memory and logic devices. The recent demonstration of room temperature stabilization of skyrmions and their current induced manipulation in industry compatible ultrathin films were first steps towards the realisation of such devices. However, important challenges remain regarding the electrical detection and the low-power nucleation of skyrmions, which are required for the read and write operations. Here, we demonstrate, using operando magnetic microscopy experiments, the electrical detection of a single magnetic skyrmion in a magnetic tunnel junction (MTJ) and its nucleation and annihilation by gate voltage via voltage control of magnetic anisotropy. The nucleated skyrmion can be further manipulated by both gate voltage and external magnetic field, leading to tunable intermediate resistance states. Our results unambiguously demonstrate the readout and voltage controlled write operations in a single MTJ device, which is a major milestone for low power skyrmion based technologies.
△ Less
Submitted 1 August, 2023;
originally announced August 2023.
-
Ext-groups in the Category of Strict Polynomial Functors
Authors:
Van Tuan Pham
Abstract:
The aim of this paper is to study, by using the mathematical tools developed by Chalupnik, Touze, and Van der Kallen, the effect of the Frobenius twist on Ext-group in the category of strict polynomial functors. As an application, we obtain explicit formulas of cohomology of the orthogonal groups and symplectic ones.
The aim of this paper is to study, by using the mathematical tools developed by Chalupnik, Touze, and Van der Kallen, the effect of the Frobenius twist on Ext-group in the category of strict polynomial functors. As an application, we obtain explicit formulas of cohomology of the orthogonal groups and symplectic ones.
△ Less
Submitted 20 June, 2023;
originally announced June 2023.
-
Ontology-based Solution for Building an Intelligent Searching System on Traffic Law Documents
Authors:
Vuong T. Pham,
Hien D. Nguyen,
Thinh Le,
Binh Nguyen,
Quoc Hung Ngo
Abstract:
In this paper, an ontology-based approach is used to organize the knowledge base of legal documents in road traffic law. This knowledge model is built by the improvement of ontology Rela-model. In addition, several searching problems on traffic law are proposed and solved based on the legal knowledge base. The intelligent search system on Vietnam road traffic law is constructed by applying the met…
▽ More
In this paper, an ontology-based approach is used to organize the knowledge base of legal documents in road traffic law. This knowledge model is built by the improvement of ontology Rela-model. In addition, several searching problems on traffic law are proposed and solved based on the legal knowledge base. The intelligent search system on Vietnam road traffic law is constructed by applying the method. The searching system can help users to find concepts and definitions in road traffic law. Moreover, it can also determine penalties and fines for violations in the traffic. The experiment results show that the system is efficient for users' typical searching and is emerging for usage in the real-world.
△ Less
Submitted 26 January, 2023;
originally announced January 2023.
-
Spin-to-charge conversion by spin pum** in sputtered polycrystalline Bi$_x$Se$_{1-x}$
Authors:
Isabel C Arango,
Alberto Anadón,
Silvestre Novoa,
Van Tuong Pham,
Won Young Choi,
Junior Alegre,
Laurent Badie,
Andrey Chuvilin,
Sébastien Petit-Watelot,
Luis E Hueso,
Fèlix Casanova,
Juan-Carlos Rojas-Sánchez
Abstract:
Topological materials are of high interest due to the promise to obtain low power and fast memory devices based on efficient spin-orbit torque switching or spin-orbit magnetic state read-out. In particular, sputtered polycrystalline Bi$_x$Se$_{1-x}$ is one of the materials with more potential for this purpose since it is relatively easy to fabricate and has been reported to have a very high spin H…
▽ More
Topological materials are of high interest due to the promise to obtain low power and fast memory devices based on efficient spin-orbit torque switching or spin-orbit magnetic state read-out. In particular, sputtered polycrystalline Bi$_x$Se$_{1-x}$ is one of the materials with more potential for this purpose since it is relatively easy to fabricate and has been reported to have a very high spin Hall angle. We study the spin-to-charge conversion in Bi$_x$Se$_{1-x}$ using the spin pum** technique coming from the ferromagnetic resonance in a contiguous permalloy thin film. We put a special emphasis on the interfacial properties of the system. Our results show that the spin Hall angle of Bi$_x$Se$_{1-x}$ has an opposite sign to the one of Pt. The charge current arising from the spin-to-charge conversion is, in contrast, lower than Pt by more than one order of magnitude. We ascribe this to the interdiffusion of Bi$_x$Se$_{1-x}$ and permalloy and the changes in chemical composition produced by this effect, which is an intrinsic characteristic of the system and is not considered in many other studies.
△ Less
Submitted 16 June, 2023; v1 submitted 24 December, 2022;
originally announced December 2022.
-
Emergence of large spin-charge interconversion at an oxidized Cu/W interface
Authors:
Inge Groen,
Van Tuong Pham,
Stefan Ilić,
Won Young Choi,
Andrey Chuvilin,
Edurne Sagasta,
Diogo C. Vaz,
Isabel C. Arango,
Nerea Ontoso,
F. Sebastian Bergeret,
Luis E. Hueso,
Ilya V. Tokatly,
Fèlix Casanova
Abstract:
Spin-orbitronic devices can integrate memory and logic by exploiting spin-charge interconversion (SCI), which is optimized by design and materials selection. In these devices, such as the magnetoelectric spin-orbit (MESO) logic, interfaces are crucial elements as they can prohibit or promote spin flow in a device as well as possess spin-orbit coupling resulting in interfacial SCI. Here, we study t…
▽ More
Spin-orbitronic devices can integrate memory and logic by exploiting spin-charge interconversion (SCI), which is optimized by design and materials selection. In these devices, such as the magnetoelectric spin-orbit (MESO) logic, interfaces are crucial elements as they can prohibit or promote spin flow in a device as well as possess spin-orbit coupling resulting in interfacial SCI. Here, we study the origin of SCI in a Py/Cu/W lateral spin valve and quantify its efficiency. An exhaustive characterization of the interface between Cu and W electrodes uncovers the presence of an oxidized layer (WO$_x$). We determine that the SCI occurs at the Cu/WO$_x$ interface with a temperature-independent interfacial spin-loss conductance of $G_{||} \approx$ 20 $\times$ 10$^{13} Ω^{-1}m^{-2}$ and an interfacial spin-charge conductivity $σ_{SC}=-$1610 $Ω^{-1}cm^{-1}$ at 10 K ($-$830 $Ω^{-1}cm^{-1}$ at 300 K). This corresponds to an efficiency given by the inverse Edelstein length $λ_{IEE}=-$0.76 nm at 10 K ($-$0.4 nm at 300 K), which is remarkably larger than in metal/metal and metal/oxide interfaces and bulk heavy metals. The large SCI efficiency at such an oxidized interface is a promising candidate for the magnetic readout in MESO logic devices.
△ Less
Submitted 16 November, 2022;
originally announced November 2022.
-
Random Utterance Concatenation Based Data Augmentation for Improving Short-video Speech Recognition
Authors:
Yist Y. Lin,
Tao Han,
Haihua Xu,
Van Tung Pham,
Yerbolat Khassanov,
Tze Yuang Chong,
Yi He,
Lu Lu,
Zejun Ma
Abstract:
One of limitations in end-to-end automatic speech recognition (ASR) framework is its performance would be compromised if train-test utterance lengths are mismatched. In this paper, we propose an on-the-fly random utterance concatenation (RUC) based data augmentation method to alleviate train-test utterance length mismatch issue for short-video ASR task. Specifically, we are motivated by observatio…
▽ More
One of limitations in end-to-end automatic speech recognition (ASR) framework is its performance would be compromised if train-test utterance lengths are mismatched. In this paper, we propose an on-the-fly random utterance concatenation (RUC) based data augmentation method to alleviate train-test utterance length mismatch issue for short-video ASR task. Specifically, we are motivated by observations that our human-transcribed training utterances tend to be much shorter for short-video spontaneous speech (~3 seconds on average), while our test utterance generated from voice activity detection front-end is much longer (~10 seconds on average). Such a mismatch can lead to suboptimal performance. Empirically, it's observed the proposed RUC method significantly improves long utterance recognition without performance drop on short one. Overall, it achieves 5.72% word error rate reduction on average for 15 languages and improved robustness to various utterance length.
△ Less
Submitted 25 May, 2023; v1 submitted 27 October, 2022;
originally announced October 2022.
-
All-electrical spin-to-charge conversion in sputtered Bi$_x$Se$_{1-x}$
Authors:
Won Young Choi,
Isabel C. Arango,
Van Tuong Pham,
Diogo C. Vaz,
Haozhe Yang,
Inge Groen,
Chia-Ching Lin,
Emily S. Kabir,
Kaan Oguz,
Punyashloka Debashis,
John J. Plombon,
Hai Li,
Dmitri E. Nikonov,
Andrey Chuvilin,
Luis E. Hueso,
Ian A. Young,
Fèlix Casanova
Abstract:
One of the major obstacles to realizing spintronic devices such as MESO logic devices is the small signal magnitude used for magnetization readout, making it important to find materials with high spin-to-charge conversion efficiency. Although intermixing at the junction of two materials is a widely occurring phenomenon, its influence on material characterization and the estimation of spin-to-charg…
▽ More
One of the major obstacles to realizing spintronic devices such as MESO logic devices is the small signal magnitude used for magnetization readout, making it important to find materials with high spin-to-charge conversion efficiency. Although intermixing at the junction of two materials is a widely occurring phenomenon, its influence on material characterization and the estimation of spin-to-charge conversion efficiencies is easily neglected or underestimated. Here, we demonstrate all electrical spin-to-charge conversion in Bi$_x$Se$_{1-x}$ nanodevices and show how the conversion efficiency can be overestimated by tens of times depending on the adjacent metal used as a contact. We attribute this to the intermixing-induced compositional change and the properties of a polycrystal that lead to drastic changes in resistivity and spin Hall angle. Strategies to improve the spin-to-charge conversion signal in similar structures for functional devices are discussed.
△ Less
Submitted 18 October, 2022;
originally announced October 2022.
-
Electrical measurement of the Spin Hall Effect isotropy in a ferromagnet
Authors:
M. Cosset-Chéneau,
M. Husien Fahmy,
A. Kandazoglou,
C. Grezes,
A. Brenac,
S. Teresi,
P. Sgarro,
P. Warin,
A. Marty,
V. T. Pham,
J. -P. Attané,
L. Vila
Abstract:
The spin-dependent transport properties of paramagnetic metals are roughly invariant under rotation. By contrast, in ferromagnetic materials the magnetization breaks the rotational symmetry, and thus the spin Hall effect is expected to become anisotropic. Here, using a specific design of lateral spin valves, we measure electrically the spin Hall Effect anisotropy in NiCu and NiPd, both in their fe…
▽ More
The spin-dependent transport properties of paramagnetic metals are roughly invariant under rotation. By contrast, in ferromagnetic materials the magnetization breaks the rotational symmetry, and thus the spin Hall effect is expected to become anisotropic. Here, using a specific design of lateral spin valves, we measure electrically the spin Hall Effect anisotropy in NiCu and NiPd, both in their ferromagnetic and paramagnetic phases. We show that the appearance of the ferromagnetic order does not lead to a sizeable anisotropy of the spin charge interconversion in these materials.
△ Less
Submitted 23 June, 2023; v1 submitted 31 May, 2022;
originally announced May 2022.
-
Skyrmions in synthetic antiferromagnets and their nucleation via electrical current and ultrafast laser illumination
Authors:
Roméo Juge,
Naveen Sisodia,
Joseba Urrestarazu Larrañaga,
Qiang Zhang,
Van Tuong Pham,
Kumari Gaurav Rana,
Brice Sarpi,
Nicolas Mille,
Stefan Stanescu,
Rachid Belkhou,
Mohamad-Assaad Mawass,
Nina Novakovic-Marinkovic,
Florian Kronast,
Markus Weigand,
Joachim Gräfe,
Sebastian Wintz,
Simone Finizio,
Jörg Raabe,
Lucia Aballe,
Michael Foerster,
Mohamed Belmeguenai,
Liliana Buda-Prejbeanu,
Justin M. Shaw,
Hans T. Nembach,
Laurent Ranno
, et al. (2 additional authors not shown)
Abstract:
Magnetic skyrmions are topological spin textures that hold great promise as nanoscale information carriers in non-volatile memory and logic devices. While room-temperature magnetic skyrmions and their current-induced manipulation were recently demonstrated, the stray field resulting from their finite magnetization as well as their topological charge limit their minimum size and reliable motion in…
▽ More
Magnetic skyrmions are topological spin textures that hold great promise as nanoscale information carriers in non-volatile memory and logic devices. While room-temperature magnetic skyrmions and their current-induced manipulation were recently demonstrated, the stray field resulting from their finite magnetization as well as their topological charge limit their minimum size and reliable motion in tracks. Antiferromagnetic (AF) skyrmions allow these limitations to be lifted owing to their vanishing magnetization and net zero topological charge, promising room-temperature, ultrasmall skyrmions, fast dynamics, and insensitivity to external magnetic fields. While room-temperature AF spin textures have been recently demonstrated, the observation and controlled nucleation of AF skyrmions operable at room temperature in industry-compatible synthetic antiferromagnetic (SAF) material systems is still lacking. Here we demonstrate that isolated skyrmions can be stabilized at zero field and room temperature in a fully compensated SAF. Using X-ray microscopy techniques, we are able to observe the skyrmions in the different SAF layers and demonstrate their antiparallel alignment. The results are substantiated by micromagnetic simulations and analytical models using experimental parameters, which confirm the chiral SAF skyrmion spin texture and allow the identification of the physical mechanisms that control the SAF skyrmion size and stability. We also demonstrate the local nucleation of SAF skyrmions via local current injection as well as ultrafast laser excitations at zero field. These results will enable the utilization of SAF skyrmions in skyrmion-based devices.
△ Less
Submitted 23 November, 2021;
originally announced November 2021.
-
Large spin-charge interconversion induced by interfacial spin-orbit coupling in a highly conducting all-metallic system
Authors:
Van Tuong Pham,
Haozhe Yang,
Won Young Choi,
Alain Marty,
Inge Groen,
Andrey Chuvilin,
F. Sebastian Bergeret,
Luis E. Hueso,
Ilya V. Tokatly,
Fèlix Casanova
Abstract:
Spin-charge interconversion in systems with spin-orbit coupling has provided a new route for the generation and detection of spin currents in functional devices for memory and logic such as spin-orbit torque switching in magnetic memories or magnetic state reading in spin-based logic. Disentangling the bulk (spin Hall effect) from the interfacial (inverse spin galvanic effect) contribution has bee…
▽ More
Spin-charge interconversion in systems with spin-orbit coupling has provided a new route for the generation and detection of spin currents in functional devices for memory and logic such as spin-orbit torque switching in magnetic memories or magnetic state reading in spin-based logic. Disentangling the bulk (spin Hall effect) from the interfacial (inverse spin galvanic effect) contribution has been a common issue to properly quantify the spin-charge interconversion in these systems, being the case of Au paradigmatic. Here, we obtain a large spin-charge interconversion at a highly conducting Au/Cu interface which is experimentally shown to arise from the inverse spin galvanic effect of the interface and not from the spin Hall effect of bulk Au. We use two parameters independent of the microscopic details to properly quantify the spin-charge interconversion and the spin losses due to the interfacial spin-orbit coupling, providing an adequate benchmarking to compare with any spin-charge interconversion system. The good performance of this metallic interface, not based in Bi, opens the path to the use of much simpler light/heavy metal systems.
△ Less
Submitted 27 June, 2024; v1 submitted 23 August, 2021;
originally announced August 2021.
-
Multitask-Based Joint Learning Approach To Robust ASR For Radio Communication Speech
Authors:
Duo Ma,
Nana Hou,
Van Tung Pham,
Haihua Xu,
Eng Siong Chng
Abstract:
To realize robust end-to-end Automatic Speech Recognition(E2E ASR) under radio communication condition, we propose a multitask-based method to joint train a Speech Enhancement (SE) module as the front-end and an E2E ASR model as the back-end in this paper. One of the advantage of the proposed method is that the entire system can be trained from scratch. Different from prior works, either component…
▽ More
To realize robust end-to-end Automatic Speech Recognition(E2E ASR) under radio communication condition, we propose a multitask-based method to joint train a Speech Enhancement (SE) module as the front-end and an E2E ASR model as the back-end in this paper. One of the advantage of the proposed method is that the entire system can be trained from scratch. Different from prior works, either component here doesn't need to perform pre-training and fine-tuning processes separately. Through analysis, we found that the success of the proposed method lies in the following aspects. Firstly, multitask learning is essential, that is the SE network is not only learning to produce more Intelligent speech, it is also aimed to generate speech that is beneficial to recognition. Secondly, we also found speech phase preserved from noisy speech is critical for improving ASR performance. Thirdly, we propose a dual channel data augmentation training method to obtain further improvement.Specifically, we combine the clean and enhanced speech to train the whole system. We evaluate the proposed method on the RATS English data set, achieving a relative WER reduction of 4.6% with the joint training method, and up to a relative WER reduction of 11.2% with the proposed data augmentation method.
△ Less
Submitted 22 July, 2021;
originally announced July 2021.
-
Disentangling spin, anomalous and planar Hall effects in ferromagnetic/heavy metal nanostructures
Authors:
Inge Groen,
Van Tuong Pham,
Naëmi Leo,
Alain Marty,
Luis E. Hueso,
Fèlix Casanova
Abstract:
Ferromagnetic (FM)/heavy metal (HM) nanostructures can be used for the magnetic state readout in the proposed magneto-electric spin-orbit logic by locally injecting a spin-polarized current and measure the spin-to-charge conversion via the spin Hall effect. However, this local configuration is prone to spurious signals. In this work, we address spurious Hall effects that can contaminate the spin H…
▽ More
Ferromagnetic (FM)/heavy metal (HM) nanostructures can be used for the magnetic state readout in the proposed magneto-electric spin-orbit logic by locally injecting a spin-polarized current and measure the spin-to-charge conversion via the spin Hall effect. However, this local configuration is prone to spurious signals. In this work, we address spurious Hall effects that can contaminate the spin Hall signal in these FM/HM T-shaped nanostructures. The most pronounced Hall effects in our Co50Fe50/Pt nanostructures are the planar Hall effect and the anomalous Hall effect generated in the FM nanowire. We find that the planar Hall effect, induced by misalignment between magnetization and current direction in the FM wire, is manifested as a shift in the measured baseline resistance, but does not alter the spin Hall signal. In contrast, the anomalous Hall effect, arising from the charge current distribution within the FM, adds to the spin Hall signal. However, the effect can be made insignificant by minimizing the shunting effect via proper design of the device. We conclude that local spin injection in FM/HM nanostructures is a suitable tool for measuring spin Hall signals and, therefore, a valid method for magnetic state readout in prospective spin-based logic.
△ Less
Submitted 6 March, 2021;
originally announced March 2021.
-
End-to-End Speaker Height and age estimation using Attention Mechanism with LSTM-RNN
Authors:
Manav Kaushik,
Van Tung Pham,
Eng Siong Chng
Abstract:
Automatic height and age estimation of speakers using acoustic features is widely used for the purpose of human-computer interaction, forensics, etc. In this work, we propose a novel approach of using attention mechanism to build an end-to-end architecture for height and age estimation. The attention mechanism is combined with Long Short-Term Memory(LSTM) encoder which is able to capture long-term…
▽ More
Automatic height and age estimation of speakers using acoustic features is widely used for the purpose of human-computer interaction, forensics, etc. In this work, we propose a novel approach of using attention mechanism to build an end-to-end architecture for height and age estimation. The attention mechanism is combined with Long Short-Term Memory(LSTM) encoder which is able to capture long-term dependencies in the input acoustic features. We modify the conventionally used Attention -- which calculates context vectors the sum of attention only across timeframes -- by introducing a modified context vector which takes into account total attention across encoder units as well, giving us a new cross-attention mechanism. Apart from this, we also investigate a multi-task learning approach for jointly estimating speaker height and age. We train and test our model on the TIMIT corpus. Our model outperforms several approaches in the literature. We achieve a root mean square error (RMSE) of 6.92cm and6.34cm for male and female heights respectively and RMSE of 7.85years and 8.75years for male and females ages respectively. By tracking the attention weights allocated to different phones, we find that Vowel phones are most important whistlestop phones are least important for the estimation task.
△ Less
Submitted 13 January, 2021;
originally announced January 2021.
-
Evidence of interfacial asymmetric spin scattering at ferromagnet-Pt interfaces
Authors:
Van Tuong Pham,
Maxen Cosset-Cheneau,
Ariel Brenac,
Olivier Boulle,
Alain Marty,
Jean-Philippe Attané,
Laurent Vila
Abstract:
We measure the spin-charge interconversion by the spin Hall effect in ferromagnetic/Pt nanodevices. The extracted effective spin Hall angles (SHAs) of Pt evolve drastically with the ferromagnetic (FM) materials (CoFe, Co, and NiFe), when assuming transparent interfaces and a bulk origin of the spin injection/detection by the FM elements. By carefully measuring the interface resistance, we show tha…
▽ More
We measure the spin-charge interconversion by the spin Hall effect in ferromagnetic/Pt nanodevices. The extracted effective spin Hall angles (SHAs) of Pt evolve drastically with the ferromagnetic (FM) materials (CoFe, Co, and NiFe), when assuming transparent interfaces and a bulk origin of the spin injection/detection by the FM elements. By carefully measuring the interface resistance, we show that it is quite large and cannot be neglected. We then evidence that the spin injection/detection at the FM/Pt interfaces are dominated by the spin polarization of the interfaces. We show that interfacial asymmetric spin scattering becomes the driving mechanism of the spin injection in our samples.
△ Less
Submitted 9 April, 2021; v1 submitted 24 November, 2020;
originally announced November 2020.
-
Enriching Under-Represented Named-Entities To Improve Speech Recognition Performance
Authors:
Tingzhi Mao,
Yerbolat Khassanov,
Van Tung Pham,
Haihua Xu,
Hao Huang,
Aishan Wumaier,
Eng Siong Chng
Abstract:
Automatic speech recognition (ASR) for under-represented named-entity (UR-NE) is challenging due to such named-entities (NE) have insufficient instances and poor contextual coverage in the training data to learn reliable estimates and representations. In this paper, we propose approaches to enriching UR-NEs to improve speech recognition performance. Specifically, our first priority is to ensure th…
▽ More
Automatic speech recognition (ASR) for under-represented named-entity (UR-NE) is challenging due to such named-entities (NE) have insufficient instances and poor contextual coverage in the training data to learn reliable estimates and representations. In this paper, we propose approaches to enriching UR-NEs to improve speech recognition performance. Specifically, our first priority is to ensure those UR-NEs to appear in the word lattice if there is any. To this end, we make exemplar utterances for those UR-NEs according to their categories (e.g. location, person, organization, etc.), ending up with an improved language model (LM) that boosts the UR-NE occurrence in the word lattice. With more UR-NEs appearing in the lattice, we then boost the recognition performance through lattice rescoring methods. We first enrich the representations of UR-NEs in a pre-trained recurrent neural network LM (RNNLM) by borrowing the embedding representations of the rich-represented NEs (RR-NEs), yielding the lattices that statistically favor the UR-NEs. Finally, we directly boost the likelihood scores of the utterances containing UR-NEs and gain further performance improvement.
△ Less
Submitted 22 October, 2020;
originally announced October 2020.
-
Imprint from ferromagnetic skyrmions in an antiferromagnet via exchange bias
Authors:
Kumari Gaurav Rana,
Rafael Lopes Seeger,
Sandra Ruiz-Gómez,
Roméo Juge,
Qiang Zhang,
Kaushik Bairagi,
Van Tuong Pham,
Mohamed Belmeguenai,
Stéphane Auffret,
Michael Foerster,
Lucia Aballe,
Gilles Gaudin,
Vincent Baltz,
Olivier Boulle
Abstract:
Magnetic skyrmions are topological spin textures holding great potential as nanoscale information carriers. Recently, skyrmions have been predicted in antiferromagnets, with key advantages in terms of stability, size, and dynamical properties over their ferromagnetic analogs. However, their experimental demonstration is still lacking. Here, we show the imprint from ferromagnetic skyrmions into a t…
▽ More
Magnetic skyrmions are topological spin textures holding great potential as nanoscale information carriers. Recently, skyrmions have been predicted in antiferromagnets, with key advantages in terms of stability, size, and dynamical properties over their ferromagnetic analogs. However, their experimental demonstration is still lacking. Here, we show the imprint from ferromagnetic skyrmions into a thin film of an IrMn antiferromagnet, at room temperature and zero external magnetic field, using exchange-bias. Using high-spatial-resolution x-ray magnetic circular dichroism photoemission electron microscopy (XMCD-PEEM), we observed the imprinted spin textures within the IrMn from the XMCD signal of the uncompensated Mn spins at the interface with the ferromagnet. This result opens up a path for logic and memory devices based on skyrmion manipulation in antiferromagnets.
△ Less
Submitted 24 November, 2021; v1 submitted 30 September, 2020;
originally announced September 2020.
-
Measurement of the Spin Absorption Anisotropy in Lateral Spin Valves
Authors:
Maxen Cosset-Chéneau,
Laurent Vila,
Gilles Zahnd,
Daria Gusakova,
Van Tuong Pham,
Cécile Grèzes,
Xavier Waintal,
Alain Marty,
Henri Jaffrès,
Jean-Philippe Attané
Abstract:
The spin absorption process in a ferromagnetic material depends on the spin orientation relativelyto the magnetization. Using a ferromagnet to absorb the pure spin current created within a lateralspin-valve, we evidence and quantify a sizeable orientation dependence of the spin absorption inCo, CoFe and NiFe. These experiments allow determining the spin-mixing conductance, an elusivebut fundamenta…
▽ More
The spin absorption process in a ferromagnetic material depends on the spin orientation relativelyto the magnetization. Using a ferromagnet to absorb the pure spin current created within a lateralspin-valve, we evidence and quantify a sizeable orientation dependence of the spin absorption inCo, CoFe and NiFe. These experiments allow determining the spin-mixing conductance, an elusivebut fundamental parameter of the spin-dependent transport. We show that the obtained valuescannot be understood within a model considering only the Larmor, transverse decoherence and spindiffusion lengths, and rather suggest that the spin-mixing conductance is actually limited by theSharvin conductance.
△ Less
Submitted 28 July, 2020;
originally announced July 2020.
-
Quantification of interfacial spin-charge conversion in metal/insulator hybrid structures by generalized boundary conditions
Authors:
Cristina Sanz-Fernández,
Van Tuong Pham,
Edurne Sagasta,
Luis E. Hueso,
Ilya V. Tokatly,
Fèlix Casanova,
F. Sebastián Bergeret
Abstract:
We present and verify experimentally a universal theoretical framework for the description of spin-charge interconversion in non-magnetic metal/insulator structures with interfacial spin-orbit coupling (ISOC). Our formulation is based on drift-diffusion equations supplemented with generalized boundary conditions. The latter encode the effects of ISOC and relate the electronic transport in such sys…
▽ More
We present and verify experimentally a universal theoretical framework for the description of spin-charge interconversion in non-magnetic metal/insulator structures with interfacial spin-orbit coupling (ISOC). Our formulation is based on drift-diffusion equations supplemented with generalized boundary conditions. The latter encode the effects of ISOC and relate the electronic transport in such systems to spin loss and spin-charge interconversion at the interface, which are parameterized, respectively, by $G_{\parallel/\perp}$ and $σ_{\rm{sc/cs}}$. We demonstrate that the conversion efficiency depends solely on these interfacial parameters. We apply our formalism to two typical spintronic devices that exploit ISOC: a lateral spin valve and a multilayer Hall bar, for which we calculate the non-local resistance and the spin Hall magnetoresistance, respectively. Finally, we perform measurements on these two devices with a BiO$_x$/Cu interface and verify that transport properties related to the ISOC are quantified by the same set of interfacial parameters.
△ Less
Submitted 3 July, 2020;
originally announced July 2020.
-
Monolingual Data Selection Analysis for English-Mandarin Hybrid Code-switching Speech Recognition
Authors:
Haobo Zhang,
Haihua Xu,
Van Tung Pham,
Hao Huang,
Eng Siong Chng
Abstract:
In this paper, we conduct data selection analysis in building an English-Mandarin code-switching (CS) speech recognition (CSSR) system, which is aimed for a real CSSR contest in China. The overall training sets have three subsets, i.e., a code-switching data set, an English (LibriSpeech) and a Mandarin data set respectively. The code-switching data are Mandarin dominated. First of all, it is found…
▽ More
In this paper, we conduct data selection analysis in building an English-Mandarin code-switching (CS) speech recognition (CSSR) system, which is aimed for a real CSSR contest in China. The overall training sets have three subsets, i.e., a code-switching data set, an English (LibriSpeech) and a Mandarin data set respectively. The code-switching data are Mandarin dominated. First of all, it is found using the overall data yields worse results, and hence data selection study is necessary. Then to exploit monolingual data, we find data matching is crucial. Mandarin data is closely matched with the Mandarin part in the code-switching data, while English data is not. However, Mandarin data only helps on those utterances that are significantly Mandarin-dominated. Besides, there is a balance point, over which more monolingual data will divert the CSSR system, degrading results. Finally, we analyze the effectiveness of combining monolingual data to train a CSSR system with the HMM-DNN hybrid framework. The CSSR system can perform within-utterance code-switch recognition, but it still has a margin with the one trained on code-switching data.
△ Less
Submitted 13 September, 2020; v1 submitted 12 June, 2020;
originally announced June 2020.
-
Leveraging Text Data Using Hybrid Transformer-LSTM Based End-to-End ASR in Transfer Learning
Authors:
Zhi** Zeng,
Van Tung Pham,
Haihua Xu,
Yerbolat Khassanov,
Eng Siong Chng,
Chongjia Ni,
Bin Ma
Abstract:
In this work, we study leveraging extra text data to improve low-resource end-to-end ASR under cross-lingual transfer learning setting. To this end, we extend our prior work [1], and propose a hybrid Transformer-LSTM based architecture. This architecture not only takes advantage of the highly effective encoding capacity of the Transformer network but also benefits from extra text data due to the L…
▽ More
In this work, we study leveraging extra text data to improve low-resource end-to-end ASR under cross-lingual transfer learning setting. To this end, we extend our prior work [1], and propose a hybrid Transformer-LSTM based architecture. This architecture not only takes advantage of the highly effective encoding capacity of the Transformer network but also benefits from extra text data due to the LSTM-based independent language model network. We conduct experiments on our in-house Malay corpus which contains limited labeled data and a large amount of extra text. Results show that the proposed architecture outperforms the previous LSTM-based architecture [1] by 24.2% relative word error rate (WER) when both are trained using limited labeled data. Starting from this, we obtain further 25.4% relative WER reduction by transfer learning from another resource-rich language. Moreover, we obtain additional 13.6% relative WER reduction by boosting the LSTM decoder of the transferred model with the extra text data. Overall, our best model outperforms the vanilla Transformer ASR by 11.9% relative WER. Last but not least, the proposed hybrid architecture offers much faster inference compared to both LSTM and Transformer architectures.
△ Less
Submitted 28 May, 2020; v1 submitted 20 May, 2020;
originally announced May 2020.
-
Approaches to Improving Recognition of Underrepresented Named Entities in Hybrid ASR Systems
Authors:
Tingzhi Mao,
Yerbolat Khassanov,
Van Tung Pham,
Haihua Xu,
Hao Huang,
Eng Siong Chng
Abstract:
In this paper, we present a series of complementary approaches to improve the recognition of underrepresented named entities (NE) in hybrid ASR systems without compromising overall word error rate performance. The underrepresented words correspond to rare or out-of-vocabulary (OOV) words in the training data, and thereby can't be modeled reliably. We begin with graphemic lexicon which allows to dr…
▽ More
In this paper, we present a series of complementary approaches to improve the recognition of underrepresented named entities (NE) in hybrid ASR systems without compromising overall word error rate performance. The underrepresented words correspond to rare or out-of-vocabulary (OOV) words in the training data, and thereby can't be modeled reliably. We begin with graphemic lexicon which allows to drop the necessity of phonetic models in hybrid ASR. We study it under different settings and demonstrate its effectiveness in dealing with underrepresented NEs. Next, we study the impact of neural language model (LM) with letter-based features derived to handle infrequent words. After that, we attempt to enrich representations of underrepresented NEs in pretrained neural LM by borrowing the embedding representations of rich-represented words. This let us gain significant performance improvement on underrepresented NE recognition. Finally, we boost the likelihood scores of utterances containing NEs in the word lattices rescored by neural LMs and gain further performance improvement. The combination of the aforementioned approaches improves NE recognition by up to 42% relatively.
△ Less
Submitted 18 May, 2020;
originally announced May 2020.
-
Spin-orbit magnetic state readout in scaled ferromagnetic/heavy metal nanostructures
Authors:
Van Tuong Pham,
Inge Groen,
Sasikanth Manipatruni,
Won Young Choi,
Dmitri E. Nikonov,
Edurne Sagasta,
Chia-Ching Lin,
Tanay Gosavi,
Alain Marty,
Luis E. Hueso,
Ian Young,
Fèlix Casanova
Abstract:
Efficient detection of the magnetic state at nanoscale dimensions is an important step to utilize spin logic devices for computing. Magnetoresistance effects have been hitherto used in magnetic state detection, but they suffer from energetically unfavorable scaling and do not generate an electromotive force that can be used to drive a circuit element for logic device applications. Here, we experim…
▽ More
Efficient detection of the magnetic state at nanoscale dimensions is an important step to utilize spin logic devices for computing. Magnetoresistance effects have been hitherto used in magnetic state detection, but they suffer from energetically unfavorable scaling and do not generate an electromotive force that can be used to drive a circuit element for logic device applications. Here, we experimentally show that a favorable miniaturization law is possible via the use of spin-Hall detection of the in-plane magnetic state of a magnet. This scaling law allows us to obtain a giant signal by spin Hall effect in CoFe/Pt nanostructures and quantify an effective spin-to-charge conversion rate for the CoFe/Pt system. The spin-to-charge conversion can be described as a current source with an internal resistance, i.e., it generates an electromotive force that can be used to drive computing circuits. We predict that the spin-orbit detection of magnetic states can reach high efficiency at reduced dimensions, paving the way for scalable spin-orbit logic devices and memories.
△ Less
Submitted 24 February, 2020;
originally announced February 2020.
-
Independent language modeling architecture for end-to-end ASR
Authors:
Van Tung Pham,
Haihua Xu,
Yerbolat Khassanov,
Zhi** Zeng,
Eng Siong Chng,
Chongjia Ni,
Bin Ma,
Haizhou Li
Abstract:
The attention-based end-to-end (E2E) automatic speech recognition (ASR) architecture allows for joint optimization of acoustic and language models within a single network. However, in a vanilla E2E ASR architecture, the decoder sub-network (subnet), which incorporates the role of the language model (LM), is conditioned on the encoder output. This means that the acoustic encoder and the language mo…
▽ More
The attention-based end-to-end (E2E) automatic speech recognition (ASR) architecture allows for joint optimization of acoustic and language models within a single network. However, in a vanilla E2E ASR architecture, the decoder sub-network (subnet), which incorporates the role of the language model (LM), is conditioned on the encoder output. This means that the acoustic encoder and the language model are entangled that doesn't allow language model to be trained separately from external text data. To address this problem, in this work, we propose a new architecture that separates the decoder subnet from the encoder output. In this way, the decoupled subnet becomes an independently trainable LM subnet, which can easily be updated using the external text data. We study two strategies for updating the new architecture. Experimental results show that, 1) the independent LM architecture benefits from external text data, achieving 9.3% and 22.8% relative character and word error rate reduction on Mandarin HKUST and English NSC datasets respectively; 2)the proposed architecture works well with external LM and can be generalized to different amount of labelled data.
△ Less
Submitted 25 November, 2019;
originally announced December 2019.
-
An Efficient Model for Sentiment Analysis of Electronic Product Reviews in Vietnamese
Authors:
Suong N. Hoang,
Linh V. Nguyen,
Tai Huynh,
Vuong T. Pham
Abstract:
In the past few years, the growth of e-commerce and digital marketing in Vietnam has generated a huge volume of opinionated data. Analyzing those data would provide enterprises with insight for better business decisions. In this work, as part of the Advosights project, we study sentiment analysis of product reviews in Vietnamese. The final solution is based on Self-attention neural networks, a fle…
▽ More
In the past few years, the growth of e-commerce and digital marketing in Vietnam has generated a huge volume of opinionated data. Analyzing those data would provide enterprises with insight for better business decisions. In this work, as part of the Advosights project, we study sentiment analysis of product reviews in Vietnamese. The final solution is based on Self-attention neural networks, a flexible architecture for text classification task with about 90.16% of accuracy in 0.0124 second, a very fast inference time.
△ Less
Submitted 29 October, 2019;
originally announced October 2019.
-
Large multi-directional spin-to-charge conversion in low symmetry semimetal MoTe$_2$ at room temperature
Authors:
C. K. Safeer,
Nerea Ontoso,
Josep Ingla-Aynés,
Franz Herling,
Van Tuong Pham,
Annika Kurzmann,
Klaus Ensslin,
Andrey Chuvilin,
Iñigo Robredo,
Maia G. Vergniory,
Fernando de Juan,
Luis E. Hueso,
M. Reyes Calvo,
Fèlix Casanova
Abstract:
Efficient and versatile spin-to-charge current conversion is crucial for the development of spintronic applications, which strongly rely on the ability to electrically generate and detect spin currents. In this context, the spin Hall effect has been widely studied in heavy metals with strong spin-orbit coupling. While the high crystal symmetry in these materials limits the conversion to the orthog…
▽ More
Efficient and versatile spin-to-charge current conversion is crucial for the development of spintronic applications, which strongly rely on the ability to electrically generate and detect spin currents. In this context, the spin Hall effect has been widely studied in heavy metals with strong spin-orbit coupling. While the high crystal symmetry in these materials limits the conversion to the orthogonal configuration, unusual configurations are expected in low symmetry transition metal dichalcogenide semimetals, which could add flexibility to the electrical injection and detection of pure spin currents. Here, we report the observation of spin-to-charge conversion in MoTe$_2$ flakes, which are stacked in graphene lateral spin valves. We detect two distinct contributions arising from the conversion of two different spin orientations. In addition to the conventional conversion where the spin polarization is orthogonal to the charge current, we also detect a conversion where the spin polarization and the charge current are parallel. Both contributions, which could arise either from bulk spin Hall effect or surface Edelstein effect, show large efficiencies comparable to the best spin Hall metals and topological insulators. Our finding enables the simultaneous conversion of spin currents with any in-plane spin polarization in one single experimental configuration.
△ Less
Submitted 10 October, 2019;
originally announced October 2019.
-
Constrained Output Embeddings for End-to-End Code-Switching Speech Recognition with Only Monolingual Data
Authors:
Yerbolat Khassanov,
Haihua Xu,
Van Tung Pham,
Zhi** Zeng,
Eng Siong Chng,
Chongjia Ni,
Bin Ma
Abstract:
The lack of code-switch training data is one of the major concerns in the development of end-to-end code-switching automatic speech recognition (ASR) models. In this work, we propose a method to train an improved end-to-end code-switching ASR using only monolingual data. Our method encourages the distributions of output token embeddings of monolingual languages to be similar, and hence, promotes t…
▽ More
The lack of code-switch training data is one of the major concerns in the development of end-to-end code-switching automatic speech recognition (ASR) models. In this work, we propose a method to train an improved end-to-end code-switching ASR using only monolingual data. Our method encourages the distributions of output token embeddings of monolingual languages to be similar, and hence, promotes the ASR model to easily code-switch between languages. Specifically, we propose to use Jensen-Shannon divergence and cosine distance based constraints. The former will enforce output embeddings of monolingual languages to possess similar distributions, while the later simply brings the centroids of two distributions to be close to each other. Experimental results demonstrate high effectiveness of the proposed method, yielding up to 4.5% absolute mixed error rate improvement on Mandarin-English code-switching ASR task.
△ Less
Submitted 31 July, 2019; v1 submitted 7 April, 2019;
originally announced April 2019.
-
Enriching Rare Word Representations in Neural Language Models by Embedding Matrix Augmentation
Authors:
Yerbolat Khassanov,
Zhi** Zeng,
Van Tung Pham,
Haihua Xu,
Eng Siong Chng
Abstract:
The neural language models (NLM) achieve strong generalization capability by learning the dense representation of words and using them to estimate probability distribution function. However, learning the representation of rare words is a challenging problem causing the NLM to produce unreliable probability estimates. To address this problem, we propose a method to enrich representations of rare wo…
▽ More
The neural language models (NLM) achieve strong generalization capability by learning the dense representation of words and using them to estimate probability distribution function. However, learning the representation of rare words is a challenging problem causing the NLM to produce unreliable probability estimates. To address this problem, we propose a method to enrich representations of rare words in pre-trained NLM and consequently improve its probability estimation performance. The proposed method augments the word embedding matrices of pre-trained NLM while kee** other parameters unchanged. Specifically, our method updates the embedding vectors of rare words using embedding vectors of other semantically and syntactically similar words. To evaluate the proposed method, we enrich the rare street names in the pre-trained NLM and use it to rescore 100-best hypotheses output from the Singapore English speech recognition system. The enriched NLM reduces the word error rate by 6% relative and improves the recognition accuracy of the rare words by 16% absolute as compared to the baseline NLM.
△ Less
Submitted 31 July, 2019; v1 submitted 7 April, 2019;
originally announced April 2019.
-
Spin dependent transport characterization in metallic lateral spin valves using 1D and 3D modeling
Authors:
P. Laczkowski,
M. Cosset-Cheneau,
W. Savero-Torres,
V. T. Pham,
H. Jaffrès,
N. Reyren,
J. -C. Rojas-Sànchez,
A. Marty,
L. Vila,
J. -M. George,
J. -P. Attané
Abstract:
We present the analysis of the spin signals obtained in NiFe based metallic lateral spin valves. We exploit the spin dependent diffusive equations in both the conventional 1D analytic modeling as well as in 3D Finite Element Method simulations. Both approaches are used for extracting the spin diffusion length $l_{sf}^{N}$ and the effective spin polarization $P_{eff}$ in Py/Al, Py/Cu and Py/Au base…
▽ More
We present the analysis of the spin signals obtained in NiFe based metallic lateral spin valves. We exploit the spin dependent diffusive equations in both the conventional 1D analytic modeling as well as in 3D Finite Element Method simulations. Both approaches are used for extracting the spin diffusion length $l_{sf}^{N}$ and the effective spin polarization $P_{eff}$ in Py/Al, Py/Cu and Py/Au based lateral nano-structures at both $300\,K$ and $77\,K$. Both the analytic modeling and 3D Finite Element Method simulations give consistent results. Combination of both models provides a powerful tool for reliable spin transport characterization in all metallic spin valves and gives an insight into the spin/charge current and spin accumulations 3D distributions in these devices. We provide the necessary ingredients to develop the 3D finite element modeling of diffusive spin transport.
△ Less
Submitted 18 March, 2019; v1 submitted 6 March, 2019;
originally announced March 2019.
-
On the End-to-End Solution to Mandarin-English Code-switching Speech Recognition
Authors:
Zhi** Zeng,
Yerbolat Khassanov,
Van Tung Pham,
Haihua Xu,
Eng Siong Chng,
Haizhou Li
Abstract:
Code-switching (CS) refers to a linguistic phenomenon where a speaker uses different languages in an utterance or between alternating utterances. In this work, we study end-to-end (E2E) approaches to the Mandarin-English code-switching speech recognition (CSSR) task. We first examine the effectiveness of using data augmentation and byte-pair encoding (BPE) subword units. More importantly, we propo…
▽ More
Code-switching (CS) refers to a linguistic phenomenon where a speaker uses different languages in an utterance or between alternating utterances. In this work, we study end-to-end (E2E) approaches to the Mandarin-English code-switching speech recognition (CSSR) task. We first examine the effectiveness of using data augmentation and byte-pair encoding (BPE) subword units. More importantly, we propose a multitask learning recipe, where a language identification task is explicitly learned in addition to the E2E speech recognition task. Furthermore, we introduce an efficient word vocabulary expansion method for language modeling to alleviate data sparsity issues under the code-switching scenario. Experimental results on the SEAME data, a Mandarin-English CS corpus, demonstrate the effectiveness of the proposed methods.
△ Less
Submitted 11 July, 2019; v1 submitted 1 November, 2018;
originally announced November 2018.
-
Cross-shaped nanostructures for the study of spin to charge interconversions using spin-orbit coupling in non-magnetic materials
Authors:
V. T. Pham,
L. Vila,
G. Zahnd,
P. Noël,
A. Marty,
J. P. Attané
Abstract:
Several spin-orbit effects allow performing spin to charge inter-conversion: the spin Hall effects, the Rashba effect, or the spin-momentum locking in topological insulators. Here we focus on how the detection of this inter-conversion can be made electrically, using three different cross-shaped nanostructures. We apply these measurement configurations to the case of the spin Hall effect in Pt, usi…
▽ More
Several spin-orbit effects allow performing spin to charge inter-conversion: the spin Hall effects, the Rashba effect, or the spin-momentum locking in topological insulators. Here we focus on how the detection of this inter-conversion can be made electrically, using three different cross-shaped nanostructures. We apply these measurement configurations to the case of the spin Hall effect in Pt, using CoFe electrodes to detect and inject spins. Both the direct and inverse spin Hall effect can be detected, with a spin Hall signal up to two order of magnitude higher than that of nonlocal measurements in metallic lateral spin valves, and with a much simpler fabrication protocol. We compare the respective signal amplitude of the three proposed geometries. Finally, comparison of the observed spin signals with finite element method calculations allows extracting the spin Hall angle and the spin diffusion length of Pt.
△ Less
Submitted 17 September, 2018;
originally announced September 2018.
-
Large enhancement of the spin Hall effect in Au by scattering with side-jump on Ta impurities
Authors:
P. Laczkowski,
Y. Fu,
H. Yang,
J. -C. Rojas-Sánchez,
P. Noel,
V. T. Pham,
G. Zahnd,
C. Deranlot,
S. Collin,
C. Bouard,
P. Warin,
V. Maurel,
M. Chshiev,
A. Marty,
J. -P. Attané,
A. Fert,
H. Jaffrès,
L. Vila,
J. -M. George
Abstract:
We present measurements of the Spin Hall Effect (SHE) in AuW and AuTa alloys for a large range of W or Ta concentrations by combining experiments on lateral spin valves and Ferromagnetic-Resonance/spin pum** technique. The main result is the identification of a large enhancement of the Spin Hall Angle (SHA) by the side-jump mechanism on Ta impurities, with a SHA as high as + 0.5 (i.e $50\%$) for…
▽ More
We present measurements of the Spin Hall Effect (SHE) in AuW and AuTa alloys for a large range of W or Ta concentrations by combining experiments on lateral spin valves and Ferromagnetic-Resonance/spin pum** technique. The main result is the identification of a large enhancement of the Spin Hall Angle (SHA) by the side-jump mechanism on Ta impurities, with a SHA as high as + 0.5 (i.e $50\%$) for about 10\% of Ta. In contrast the SHA in AuW does not exceed + 0.15 and can be explained by intrinsic SHE of the alloy without significant extrinsic contribution from skew or side-jump scattering by W impurities. The AuTa alloys, as they combine a very large SHA with a moderate resistivity (smaller than $85\,μΩ.cm$), are promising for spintronic devices exploiting the SHE.
△ Less
Submitted 30 August, 2017;
originally announced August 2017.
-
Fixed point forms of the parallel symmetric sandpile model
Authors:
E. Formenti,
V. T. Pham,
H. D. Phan,
T. T. H. Tran
Abstract:
This paper presents a generalization of the sandpile model, called the parallel symmetric sandpile model, which inherits the rules of the symmetric sandpile model and implements them in parallel. In this new model, at each step the collapsing of the collapsible columns happens at the same time and one collapsible column is able to collapse on the left or on the right but not both. We prove that th…
▽ More
This paper presents a generalization of the sandpile model, called the parallel symmetric sandpile model, which inherits the rules of the symmetric sandpile model and implements them in parallel. In this new model, at each step the collapsing of the collapsible columns happens at the same time and one collapsible column is able to collapse on the left or on the right but not both. We prove that the set of forms of fixed points of the symmetric sandpile model is the same as the one of that model using parallel update scheme by constructing explicitly the way (in the parallel update scheme) to reach the form of an arbitrary fixed point of the sequential model.
△ Less
Submitted 2 July, 2012; v1 submitted 5 September, 2011;
originally announced September 2011.
