arXiv subjects

Shuhang Wu

Publications and source records attributed to Shuhang Wu.

5 recordsLinked to original sources

arXiv2026

SpeakerRPL v2: Robust Open-set Speaker Identification through Enhanced Few-shot Foundation Tuning and Model Fusion

This paper proposes an improved approach for open-set speaker identification based on pretrained speaker foundation models. Building upon the previous Speaker Reciprocal Points Learning framework (V1), we first introduce an enhanced open-set learning objective by integrating reciprocal points learning with logit normalization (LogitNorm) and incorporating adaptive anchor learning to better constrain target speaker representations and improve robustness. Second, we propose a model fusion strategy to stabilize and enhance the few-shot tuning process, effectively reducing result randomness and improving generalization. Furthermore, we introduce a model selection method to ensure optimal performance in model fusion. Experimental evaluations on the VoxCeleb, ESD and 3D-Speaker datasets demonstrate the effectiveness and robustness of the proposed method under diverse conditions. On a newly proposed Vox1-O-like test set, our method reduces the EER from 1.28% to 0.09%, achieving a relative reduction of approximately 93%.

eess.AS↗

arXiv2026

The USTC-NERCSLIP Systems for the CHiME-9 MCoRec Challenge

This report details our submission to the CHiME-9 MCoRec Challenge on recognizing and clustering multiple concurrent natural conversations within indoor social settings. Unlike conventional meetings centered on a single shared topic, this scenario contains multiple parallel dialogues--up to eight speakers across up to four simultaneous conversations--with a speech overlap rate exceeding 90%. To tackle this, we propose a multimodal cascaded system that leverages per-speaker visual streams extracted from synchronized 360 degree video together with single-channel audio. Our system improves three components of the pipeline by leveraging enhanced audio-visual pretrained models: Active Speaker Detection (ASD), Audio-Visual Target Speech Extraction (AVTSE), and Audio-Visual Speech Recognition (AVSR). The AVSR module further incorporates Whisper and LLM techniques to boost transcription accuracy. Our best single cascaded system achieves a Speaker Word Error Rate (WER) of 32.44% on the development set. By further applying ROVER to fuse outputs from diverse front-end and back-end variants, we reduce Speaker WER to 31.40%. Notably, our LLM-based zero-shot conversational clustering achieves a speaker clustering F1 score of 1.0, yielding a final Joint ASR-Clustering Error Rate (JACER) of 15.70%.

eess.AS↗

arXiv2025

Selective Oxidation and Cr Segregation in High-Entropy Oxide Thin Films

High-entropy oxides (HEOs) offer exceptional compositional flexibility and structural stability, making them promising materials for energy and catalytic applications. Here, we investigate Sr doping effects on B-site cation oxidation states, local composition, and structure in epitaxial La1-xSrx(Cr0.2Mn0.2Fe0.2Co0.2Ni0.2)O3 thin films. X-ray spectroscopies reveal that Sr doping preferentially promotes Cr oxidation from Cr3+ to Cr6+, partially oxidizes Co and Ni, while leaving Mn4+ and Fe3+ unchanged. Atomic-resolution scanning transmission electron microscopy with energy-dispersive X-ray spectroscopy shows pronounced Cr segregation, with Cr exhibiting depletion at the film-substrate interface and enrichment at the film surface, along with the formation of a partially amorphous phase in heavily Sr-doped samples. This segregation is likely driven by oxidation-induced migration of smaller, high-valence Cr cations during the growth. These findings underscore the critical interplay between charge transfer, local strain, and compositional fluctuations, providing strategies to control surface composition and electronic structure in HEOs for more robust electrocatalyst design.

cond-mat.mtrl-sci↗

arXiv2015

Asymptotic Error Free Partitioning over Noisy Boolean Multiaccess Channels

In this paper, we consider the problem of partitioning active users in a manner that facilitates multi-access without collision. The setting is of a noisy, synchronous, Boolean, multi-access channel where $K$ active users (out of a total of $N$ users) seek to access. A solution to the partition problem places each of the $N$ users in one of $K$ groups (or blocks) such that no two active nodes are in the same block. We consider a simple, but non-trivial and illustrative case of $K=2$ active users and study the number of steps $T$ used to solve the partition problem. By random coding and a suboptimal decoding scheme, we show that for any $T\geq (C_1 +ξ_1)\log N$, where $C_1$ and $ξ_1$ are positive constants (independent of $N$), and $ξ_1$ can be arbitrary small, the partition problem can be solved with error probability $P_e^{(N)} \to 0$, for large $N$. Under the same scheme, we also bound $T$ from the other direction, establishing that, for any $T \leq (C_2 - ξ_2) \log N$, the error probability $P_e^{(N)} \to 1$ for large $N$; again $C_2$ and $ξ_2$ are constants and $ξ_2$ can be arbitrarily small. These bounds on the number of steps are lower than the tight achievable lower-bound in terms of $T \geq (C_g +ξ)\log N $ for group testing (in which all active users are identified, rather than just partitioned). Thus, partitioning may prove to be a more efficient approach for multi-access than group testing.

cs.IT↗

arXiv2014

Partition Information and its Transmission over Boolean Multi-Access Channels

In this paper, we propose a novel partition reservation system to study the partition information and its transmission over a noise-free Boolean multi-access channel. The objective of transmission is not message restoration, but to partition active users into distinct groups so that they can, subsequently, transmit their messages without collision. We first calculate (by mutual information) the amount of information needed for the partitioning without channel effects, and then propose two different coding schemes to obtain achievable transmission rates over the channel. The first one is the brute force method, where the codebook design is based on centralized source coding; the second method uses random coding where the codebook is generated randomly and optimal Bayesian decoding is employed to reconstruct the partition. Both methods shed light on the internal structure of the partition problem. A novel hypergraph formulation is proposed for the random coding scheme, which intuitively describes the information in terms of a strong coloring of a hypergraph induced by a sequence of channel operations and interactions between active users. An extended Fibonacci structure is found for a simple, but non-trivial, case with two active users. A comparison between these methods and group testing is conducted to demonstrate the uniqueness of our problem.

cs.IT↗