Changhao Cheng | Searcharxiv

arXiv2026

On the Distillation Loss Functions of Speech VAE for Unified Reconstruction, Understanding, and Generation

Continuous speech representations based on Variational Autoencoders (VAEs) have emerged as a promising alternative to traditional spectrogram or discrete token based features for speech generation and reconstruction. Recent research has tried to enrich the structural information in VAE latent representations by aligning with self-supervised learning (SSL) features, aiming for better generation performance. However, it remains unclear whether the widely-used alignment approach based on time-axis distillation is optimal when considering more tasks. To address this problem, this paper systematically explores different alignment approaches and analyzes their impact on the performances over three axes: reconstruction, understanding, and generation. We investigate various design choices in the distillation loss. Extensive experiments show that the joint-marginal alignment approach with adaptive weighting can achieve the best overall performance while allowing for a controllable balance.

cs.SD↗

arXiv2025

Thermal-Gradient Cooling of Atomic Vapor Fluid

The pursuit of high optical depth and long coherence time in atomic ensembles faces a fundamental thermodynamic constraint: heating enhances light-atom coupling via increased density but degrades coherence through thermal broadening, while laser cooling preserves coherence at the cost of density loss. Here, we demonstrate a non-equilibrium strategy that spatially achieves a negative correlation between density and temperature via controlled thermal-gradient transport. By engineering a temperature gradient via laser-cooling in a hot vapor cell, we drive a convective atomic fluid that expels hot atoms at the boundary while confining low-temperature atoms in the central region. This dynamic process sustains a density of 10^22m^-3 and a temperature of tens of kelvins at the center. A theoretical scheme based on the Boltzmann-type transport equation is established, which gives Navier-Stokes equations for non-equilibrium thermal-gradient atomic fluid. The results of numerical simulation indicate that this scheme can enhance the optical depth while reducing the temperature of the system, establishing a route to bypass equilibrium thermodynamics in room-temperature atom-light interactions, boosting high-performance quantum metrology and quantum information applications.

physics.atom-ph↗