arXiv subjects

Daren Chen

Publications and source records attributed to Daren Chen.

15 recordsLinked to original sources

arXiv2026

Qwen-Audio-3.0-ASR Technical Report

In recent years, automatic speech recognition (ASR) has witnessed transformative advancements driven by three complementary paradigms: data scaling, model scaling, and deep integration with large language models (LLMs). However, bridging the gap between academic benchmark performance and real-world production utility remains a persistent challenge, particularly in handling diverse regional dialects, dynamic entities and hotwords, long-range contextual information, and disfluent spontaneous speech. In this report, we present Qwen-Audio-3.0-ASR, a Mixture-of-Experts (MoE) LLM-based ASR system designed to address these production demands through a unified, instruction-following framework. The model is built upon the Qwen backbone, and is trained on tens of millions of hours of large-scale speech data. Qwen-Audio-3.0-ASR supports transcription across 30 languages and 16 Chinese dialectal varieties spanning eight major dialect regions. Beyond multilingual and dialectal recognition, the model provides production-oriented capabilities including industry-domain entity recognition, hierarchical hotword customization, native single-pass transcription polishing, and long-audio contextual modeling. We further develop a dedicated streaming variant, Qwen-Audio-3.0-ASR-Streaming, for latency-sensitive applications. Extensive evaluations on Chinese, English, multilingual, and real-world industrial test sets demonstrate state-of-the-art or highly competitive recognition performance across a broad range of evaluation conditions, with strong performance relative to leading commercial and proprietary systems including GPT-4o Transcribe and Gemini 3.1 Pro.

cs.CL↗

arXiv2026

Telecom quantum memory over one microsecond in nanophotonic lithium niobate

Nanophotonic quantum memory is a vital component for scalable quantum information processing for quantum computing, networking, and sensing applications. We store single-photon-level telecom-band optical pulses for more than a microsecond using an atomic frequency comb in erbium-doped thin-film lithium niobate, well beyond what is practically feasible via propagation in even the best nanophotonic devices due to propagation losses. We verify the quantum nature of this storage by demonstrating the phase coherence and sub-single-photon noise upon retrieval. We also show the flexibility of our platform by storing up to 20 temporal modes and demonstrating an acceptance bandwidth up to 2.2 GHz. These results establish erbium-doped thin-film lithium niobate as a practical platform for on-chip quantum memory at telecom wavelengths, a key missing element for photonic quantum computing and quantum networking.

quant-ph↗

arXiv2026

FlowSE-GRPO: Training Flow Matching Speech Enhancement via Online Reinforcement Learning

Generative speech enhancement offers a promising alternative to traditional discriminative methods by modeling the distribution of clean speech conditioned on noisy inputs. Post-training alignment via reinforcement learning (RL) effectively aligns generative models with human preferences and downstream metrics in domains such as natural language processing, but its use in speech enhancement remains limited, especially for online RL. Prior work explores offline methods like Direct Preference Optimization (DPO); online methods such as Group Relative Policy Optimization (GRPO) remain largely uninvestigated. In this paper, we present the first successful integration of online GRPO into a flow-matching speech enhancement framework, enabling efficient post-training alignment to perceptual and task-oriented metrics with few update steps. Unlike prior GRPO work on Large Language Models, we adapt the algorithm to the continuous, time-series nature of speech and to the dynamics of flow-matching generative models. We show that optimizing a single reward yields rapid metric gains but often induces reward hacking that degrades audio fidelity despite higher scores. To mitigate this, we propose a multi-metric reward optimization strategy that balances competing objectives, substantially reducing overfitting and improving overall performance. Our experiments validate online GRPO for speech enhancement and provide practical guidance for RL-based post-training of generative audio models.

eess.AS↗

arXiv2026

E2E-AEC: Implementing an end-to-end neural network learning approach for acoustic echo cancellation

We propose a novel neural network-based end-to-end acoustic echo cancellation (E2E-AEC) method capable of streaming inference, which operates effectively without reliance on traditional linear AEC (LAEC) techniques and time delay estimation. Our approach includes several key strategies: First, we introduce and refine progressive learning to gradually enhance echo suppression. Second, our model employs knowledge transfer by initializing with a pre-trained LAECbased model, harnessing the insights gained from LAEC training. Third, we optimize the attention mechanism with a loss function applied on attention weights to achieve precise time alignment between the reference and microphone signals. Lastly, we incorporate voice activity detection to enhance speech quality and improve echo removal by masking the network output when near-end speech is absent. The effectiveness of our approach is validated through experiments conducted on public datasets.

cs.SD↗

arXiv2026

The link surgery modules of 2-component L-space links

In our earlier work, we studied the link surgery modules of two component L-space links. Therein, we computed two of the four idempotents of such modules. In this article, we use Koszul duality to give an alternate account of this proof, and also to extend it to compute the entire link surgery modules of such links, modulo a technical result which will be proven in a subsequent paper.

math.GT↗

arXiv2025

Applications of the L-space satellite formula

We give a formula for the $\tau$-invariant of a satellite knot $P(K,n)$ when $P$ is an L-space satellite operator. Our formula holds for general L-space satellite operators $P$ when the companion $K$ satisfies $\epsilon(K)=1$. When $\epsilon(K)$ is $0$ or $-1$, we state a formula which requires some additional assumptions on $P$ or $n$. Our main tool is our algorithm which computes the knot Floer complex of satellite knots constructed using L-space satellite operators, which we developed in a previous paper. Our formula for $\tau$ recovers many existing formulas for the behavior of $\tau$ under satellite operators, including for cables. We apply our formula to questions about the slice genus of satellite knots, showing, e.g., that if $K$ is a knot with $\tau(K)=g_4(K)>0$, then satellites of $K$ by L-space satellite operators have the same property. Another application is a proof that L-space satellite operators satisfy a conjecture of Hedden and Pinz\'on-Caicedo: If $P$ is an L-space satellite operator which acts as a group homomorphism on the smooth concordance group, then $P$ is either the zero operator, the identity operator, or the orientation reversing operator.

math.GT↗

arXiv2025

The flip map and involutions on Khovanov homology

The flip symmetry on knot diagrams induces an involution on Khovanov homology. We prove that this involution is determined by its behavior on unlinks; in particular, it is the identity map when working over $\mathbb{F}_2$. This confirms a folklore conjecture on the triviality of the Viro flip map. As a corollary, we prove that the symmetries on the transvergent and intravergent diagrams of a strongly invertible knot induce the same involution on Khovanov homology. We also apply similar techniques to study the half sweep-around map.

math.GT↗

arXiv2024

L-space satellite operators and knot Floer homology

We consider satellite operators where the corresponding 2-component link is an L-space link. This family includes many commonly studied satellite operators, including cabling operators, the Whitehead operator, and a family of Mazur operators. We give a formula which computes the knot Floer complex of a satellite of $K$ in terms of the knot Floer complex of $K$. Our main tools are the Heegaard Floer Dehn surgery formulas and their refinements. A key step in our computation is a proof that 2-component L-space links have formal knot Floer complexes. We use this to show that the link Floer complexes of 2-component L-space links are determined by their multivariable Alexander polynomials. We implement our satellite formula in Python code, which we also make available.

math.GT↗

arXiv2023

A note on rational band moves

We introduce an oriented rational band move, a generalization of an ordinary oriented band move, and show that if a knot $K$ in the three-sphere can be made into the $(n+1)$-component unlink by $n$ oriented rational band moves, then $K$ is rationally slice.

math.GT↗

arXiv2023

Bar-Natan Homology for null homologous links in \mathbb{RP}^3

In this paper, we introduce Bar-Natan homology for null homologous links in \mathbb{RP}^3 over the field of two elements. It is a deformation of the Khovanov homology in \mathbb{RP}^3 defined by Asaeda, Przytycki and Sikora. We also define an s-invariant from this deformation using the same recipe as for links in S^3, and prove some genus bound using it. The key ingredient is the notion of twisted orientation for null homologous links and cobordisms in \mathbb{RP}^3.

math.GT↗

arXiv2023

Adiabatic compactness for holomorphic curves with boundary on nearby Lagrangians

In his 1989 paper, Floer established a connection between holomorphic strips with boundary on a Lagrangian $L$ and a small Hamiltonian push-off $L_{f}$, and gradient flow lines for the function $f$. The present paper studies the compactness theory for holomorphic curves $u_{n}$ whose boundary components lie on Hamiltonian perturbations $L_{n}^{1},\dots,L^{N}_{n}$ of a fixed Lagrangian $L$, where each sequence of nearby Lagrangians $L^{j}_{n}$ converges to $L$ as $n\to\infty$. Generalizing earlier work of Oh, Fukaya, Ekholm, and Zhu, we prove that the limit of a sequence of such holomorphic maps is a configuration consisting of holomorphic curves with boundary on $L$ joined by gradient flow lines connecting points on the boundary of holomorphic pieces. The key new result is an exponential estimate analyzing the interface between the holomorphic parts and the gradient flow line parts.

math.SG↗

arXiv2022

Floer lasagna modules from link Floer homology

In this paper, we introduce the notion of Floer lasagna modules, which is inspired by the construction of skein lasagna module in [MWW19] by Morrison, Walker and Wedrich. Here we use link Floer homology instead of Khovanov-Rozansky homology. We give a description of the Floer lasagna module for 4-manifolds obtained by adding 2-handles to the 4-ball and we compute some examples.

math.GT↗

arXiv2022

Twistings and the Alexander polynomial

We give an explicit formula of the Alexander polynomial of the link obtained by adding an arbitrary number of full twists to positively oriented parallel n-strands in terms of the Alexander polynomials of the links obtained by adding 0,1,...,n-1 full twists. From this, we see that the Alexander polynomials stabilize after adding sufficiently many full twists. The main tool used in the computation is expressing the Alexander polynomial using the vector space representation of $U_{q}(gl(1|1))$.

math.GT↗

arXiv2021

Generalised homomorphisms, measuring coalgebras and extended symmetries

Three categories of algebras with morphisms generalising the usual set of algebra homomorphisms are described. The Sweedler product provides a hom-tensor equivalence relating these three categories, and a tool enabling the universal measuring coalgebra to be calculated in small cases. A parallel theory for modules is presented.

math.RA↗

arXiv2021

Khovanov-type homologies of null homologous links in $\mathbb{RP}^3$

Let L be a null homologous link in $\mathbb{RP}^3$. We define Khovanov-type homologies of L which depend on an extra input $α= (V_0,V_1,f,g)$ consisting of two graded vectors spaces and two maps between them. With some specific choice of $α= α_{APS}$, we recover the categorification of the Kauffman bracket due to Asaeda-Przytycki-Sikora. With another choice of $α= α_{HF}$, we construct a spectral sequence from our theory converging to the Heegaard Floer homology of the even branched double cover of $\mathbb{RP}^3$.

math.GT↗