arXiv subjects

Jianjin Wang

Publications and source records attributed to Jianjin Wang.

9 recordsLinked to original sources

arXiv2026

On the Emotion Understanding of Synthesized Speech

Emotion is a core paralinguistic feature in voice interaction. It is widely believed that emotion understanding models learn fundamental representations that transfer to synthesized speech, making emotion understanding results a plausible reward or evaluation metric for assessing emotional expressiveness in speech synthesis. In this work, we critically examine this assumption by systematically evaluating Speech Emotion Recognition (SER) on synthesized speech across datasets, discriminative and generative SER models, and diverse synthesis models. We find that current SER models can not generalize to synthesized speech, largely because speech token prediction during synthesis induces a representation mismatch between synthesized and human speech. Moreover, generative Speech Language Models (SLMs) tend to infer emotion from textual semantics while ignoring paralinguistic cues. Overall, our findings suggest that existing SER models often exploit non-robust shortcuts rather than capturing fundamental features, and paralinguistic understanding in SLMs remains challenging.

cs.CL↗

arXiv2025

MTP-S2UT: Enhancing Speech-to-Speech Translation Quality with Multi-token Prediction

Current direct speech-to-speech translation methods predominantly employ speech tokens as intermediate representations. However, a single speech token is not dense in semantics, so we generally need multiple tokens to express a complete semantic unit. To address this limitation, we introduce multi-token prediction (MTP) loss into speech-to-unit translation (S2UT) models, enabling models to predict multiple subsequent tokens at each position, thereby capturing more complete semantics and enhancing information density per position. Initial MTP implementations apply the loss at the final layer, which improves output representation but initiates information enrichment too late. We hypothesize that advancing the information enrichment process to intermediate layers can achieve earlier and more effective enhancement of hidden representation. Consequently, we propose MTP-S2UT loss, applying MTP loss to hidden representation where CTC loss is computed. Experiments demonstrate that all MTP loss variants consistently improve the quality of S2UT translation, with MTP-S2UT achieving the best performance.

cs.CL↗

arXiv2024

A Modular-based Strategy for Mitigating Gradient Conflicts in Simultaneous Speech Translation

Simultaneous Speech Translation (SimulST) involves generating target language text while continuously processing streaming speech input, presenting significant real-time challenges. Multi-task learning is often employed to enhance SimulST performance but introduces optimization conflicts between primary and auxiliary tasks, potentially compromising overall efficiency. The existing model-level conflict resolution methods are not well-suited for this task which exacerbates inefficiencies and leads to high GPU memory consumption. To address these challenges, we propose a Modular Gradient Conflict Mitigation (MGCM) strategy that detects conflicts at a finer-grained modular level and resolves them utilizing gradient projection. Experimental results demonstrate that MGCM significantly improves SimulST performance, particularly under medium and high latency conditions, achieving a 0.68 BLEU score gain in offline tasks. Additionally, MGCM reduces GPU memory consumption by over 95\% compared to other conflict mitigation methods, establishing it as a robust solution for SimulST tasks.

cs.CL↗

arXiv2024

Antipersistent energy current correlations in strong long-ranged Fermi-Pasta-Ulam-Tsingou type models

We study heat transfer in one-dimensional Fermi-Pasta-Ulam-Tsingou type systems with long-range (LR) interactions. The strength of the LR interaction between two lattice sites decays as a power $σ$ of the inverse of their distance. We focus on the strong LR regime ($0\leq σ\leq1$) and show that the thermal transport behaviors are remarkably nuanced. Specifically, we observe that the antipersistent (negative) energy current correlation in this regime is intricately dependent on $σ$, displaying a nonmonotonic variation. Notably, a significant qualitative change occurs at $σ_c=0.5$, where with respect to other $σ$ values, the correlation shows a minimum negative value. Furthermore, our findings also demonstrate that within the long-time range considered, these antipersistent correlations will eventually vanish for certain $σ>0.5$. The underlying mechanisms behind these intriguing phenomena are related to the crossover of two diverse space-time scaling properties of equilibrium heat correlations and the various scattering processes of phonons and discrete breathers.

cond-mat.stat-mech↗

arXiv2024

Observation of Ballistic Thermal Transport in a Nonintegrable Classical Many-Body System

We report, for the first time, the observation of ballistic thermal transport in a nonintegrable classical many-body system. This claim is substantiated by appropriately incorporating long-range interactions into the system, which exhibits all characteristic hallmarks of ballistic heat transport, including the presence of equilibrium dynamical correlations exhibiting ballistic scaling, a size-independent energy current and a flat bulk temperature profile. These findings hold true for large system sizes (long times), indicating that ballistic heat transport is valid in the thermodynamic limit. The underlying mechanism is attributed to the presence of traveling discrete breathers in the relevant nonintegrabel systems surpassing conventional solitons in a nonlinear integrable Toda system.

cond-mat.stat-mech↗

arXiv2023

Thermal conductivity in one-dimensional nonlinear disordered lattices: Two kinds of scattering effects of hard-type and soft-type anharmonicities

The amorphous solids can be theoretically modeled by anharmonic disordered lattices. However, most of theoretical studies on thermal conductivity in anharmonic disordered lattices only focus on the potentials of hard-type (HT) anharmonicity. Here we study the thermal conductivity $κ$ of one-dimensional (1D) disordered lattices with both hard- and soft-type (ST) anharmonic on-site potentials. It is found, via both direct molecular dynamic simulations and theoretical method, that the anharmonicity dependence of $κ$ in the HT model is nonmonotonous, while in the ST model is monotonously increased. This provides a new way to enhance thermal conductivity in disordered systems. Furthermore, $κ$ of the HT model is consistent with the prediction of the quasi-harmonic Green-Kubo (QHGK) method in a wide range of anharmonicity, while for the ST model, the numerical results seem largely deviated from the theoretical predictions as the anharmonicity becomes soft. This new and peculiar feature of the ST model may root in the fact that only delocalization effect exists, different from the competing roles that both delocalization and localization play in the counterpart HT model.

cond-mat.stat-mech↗

arXiv2021

Subdiffusive Energy Transport and Antipersistent Correlations Due to the Scattering of Phonons and Discrete Breathers

While there are many physical processes showing subdiffusion and some useful particle models for understanding the underlying mechanisms have been established, a systematic study of subdiffusive energy transport is still lacking. Here we present convincing evidence that the energy subdiffusion and its antipersistent correlations take place in a Hamiltonian lattice system with both harmonic nearest-neighbor and anharmonic long-range interactions. We further understand the underlying mechanisms from the scattering of phonons and discrete breathers. Our result sheds new light on understanding the extremely slow energy transport.

cond-mat.stat-mech↗

arXiv2020

Thermal transport in long-range interacting Fermi-Pasta-Ulam chains

Studies of thermal transport in long-range (LR)interacting systems are currently particularly challenging. The main difficulties lie in the choice of boundary conditions and the definition of heat current when driving systems in an out-of-equilibrium state by the usual thermal reservoirs. Here, by employing a reverse type of thermal baths that can overcome such difficulties, we reveal the intrinsic features of thermal transport underlying a LR interacting Fermi-Pasta-Ulam chain. We find that under an appropriate range value of LR exponent $σ=2$, while a \emph{nonballistic} power-law length ($L$) divergence of thermal conductivity $κ$, i.e., $κ\sim L^α$ still persists, its scaling exponent $α\simeq 0.7$ can be much larger than the usual predictions in short-range interacting systems. The underlying mechanism is related to the system's new heat diffusion process, weaker nonintegrability and peculiar dynamics of traveling discrete breathers. Our results shed light on searching for low-dimensional materials supporting higher thermal conductivity by involving appropriate LR interactions.

cond-mat.stat-mech↗

arXiv2015

Non-Gaussian normal diffusion induced by delocalization

The non-Gaussian normal diffusion, i.e., the probability distribution function (PDF) is non-Gaussian but the mean squared displacement (MSD) depends on time linearly, has been observed in particle motions. Here we show by numerical simulations that this phenomenon may manifest in energy diffusion along lattices at a non-zero, finite temperature. The model we study is one-dimensional disordered lattices with on-site potential. We find that the energy-density fluctuations are spatially localized if the nonlinear interaction is suppressed, but may relax with a non-Gaussian PDF and a linear time-dependent MSD when the nonlinear interaction is turned on. Our analysis suggests that the mechanism lies in the delocalization properties of the localized modes.

cond-mat.dis-nn↗