arXiv subjects

Xiulian Peng

Publications and source records attributed to Xiulian Peng.

At least 19 recordsLinked to original sources

arXiv2026

Closed-Loop Triplet Synergistic Generation for Long-Form Video

Multi-shot long-form video generation remains challenging due to identity drift and compounding inconsistencies across shots. While storyboard-driven pipelines improve controllability, they are often executed in a feed-forward manner, with limited mechanisms to incorporate generated visual evidence back into subsequent conditioning. We propose CoTriSyGen, an agentic framework that formulates multi-shot long video generation as a closed-loop visual-text-memory synergy process, where planned intent, persistent memory, and generated visuals are jointly leveraged for iterative correction and long-range coherence. A vision-language-model-based analyzer reasons over this triplet and produces updates to both prompts and memory along two pathways: (i) intra-shot refinement, which triggers targeted regeneration when semantic or compositional violations are detected and refines image-to-video prompt for coherent motions; and (ii) inter-shot refinement, which rewrites subsequent-shot prompts to propagate newly manifested entities or attributes and improve prompt quality (e.g., compositional grounding and cinematic fluency) based on generated evidence. The loop is grounded in an entity-centric memory modeled as a mutable visual state that evolves as the story progresses, which is continuously updated by both the generator and the analyzer by adding new and evolved entities to reflect appearance changes, accumulated multi-view evidence, and multi-entity compositions. Experiments on our curated StoryBench benchmark demonstrate substantial improvements in cross-shot consistency, prompt adherence, and cinematic continuity over representative methods.

cs.CV↗

arXiv2026

From Content to Knowledge: Lightning Fast Long-Video Understanding with Neural Knowledge Representations

We propose a new paradigm for long video understanding by treating a long video as a Neural Knowledge Representation (NKR). NKR represents video contents neither as a stream of tokens nor pre-organized databases, but as an individual small portion of network weights attached to the VLM backbone. The NKR weights are optimized to encapsulate the video's semantic content via a novel Agentic Knowledge Distillation (AKD) process, where an agent automatically synthesizes dense descriptions and question-answer pairs to distill the video's knowledge into the NKR. While AKD serves as a comprehensive, one-time encoding phase, the resulting NKR transforms the video into a portable, reusable asset. At inference, the lightweight NKR is mounted onto a frozen Vision-Language Model (VLM), enabling direct, query-based understanding without reloading or re-encoding the original video. This approach decouples video length from inference cost, offering high amortized efficiency for multi-turn video understanding. Experiments on the LVBench benchmark show our method achieves performance comparable to state-of-the-art approaches while reducing end-to-end latency by over two orders of magnitude, opening new possibilities for interactive long-video understanding.

cs.CV↗

arXiv2026

Hierarchical Long Video Understanding with Audiovisual Entity Cohesion and Agentic Search

Long video understanding presents significant challenges for vision-language models due to extremely long context windows. Existing solutions relying on naive chunking strategies with retrieval-augmented generation, typically suffer from information fragmentation and a loss of global coherence. We present HAVEN, a unified framework for long-video understanding that enables coherent and comprehensive reasoning by integrating audiovisual entity cohesion and hierarchical video indexing with agentic search. First, we preserve semantic consistency by integrating entity-level representations across visual and auditory streams, while organizing content into a structured hierarchy spanning global summary, scene, segment, and entity levels. Then we employ an agentic search mechanism to enable dynamic retrieval and reasoning across these layers, facilitating coherent narrative reconstruction and fine-grained entity tracking. Extensive experiments demonstrate that our method achieves good temporal coherence, entity consistency, and retrieval efficiency, establishing a new state-of-the-art with an overall accuracy of 84.1% on LVBench. Notably, it achieves outstanding performance in the challenging reasoning category, reaching 80.1%. These results highlight the effectiveness of structured, multimodal reasoning for comprehensive and context-consistent understanding of long-form videos.

cs.CV↗

arXiv2025

Text-Queried Audio Source Separation via Hierarchical Modeling

Target audio source separation with natural language queries presents a promising paradigm for extracting arbitrary audio events through arbitrary text descriptions. Existing methods mainly face two challenges, the difficulty in jointly modeling acoustic-textual alignment and semantic-aware separation within a blindly-learned single-stage architecture, and the reliance on large-scale accurately-labeled training data to compensate for inefficient cross-modal learning and separation. To address these challenges, we propose a hierarchical decomposition framework, HSM-TSS, that decouples the task into global-local semantic-guided feature separation and structure-preserving acoustic reconstruction. Our approach introduces a dual-stage mechanism for semantic separation, operating on distinct global and local semantic feature spaces. We first perform global-semantic separation through a global semantic feature space aligned with text queries. A Q-Audio architecture is employed to align audio and text modalities, serving as pretrained global-semantic encoders. Conditioned on the predicted global feature, we then perform the second-stage local-semantic separation on AudioMAE features that preserve time-frequency structures, followed by acoustic reconstruction. We also propose an instruction processing pipeline to parse arbitrary text queries into structured operations, extraction or removal, coupled with audio descriptions, enabling flexible sound manipulation. Our method achieves state-of-the-art separation performance with data-efficient training while maintaining superior semantic consistency with queries in complex auditory scenes.

cs.SD↗

arXiv2025

Universal Speech Token Learning via Low-Bitrate Neural Codec and Pretrained Representations

Current large speech language models are mainly based on semantic tokens from discretization of self-supervised learned representations and acoustic tokens from a neural codec, following a semantic-modeling and acoustic-synthesis paradigm. However, semantic tokens discard paralinguistic attributes of speakers that is important for natural spoken communication, while prompt-based acoustic synthesis from semantic tokens has limits in recovering paralinguistic details and suffers from robustness issues, especially when there are domain gaps between the prompt and the target. This paper unifies two types of tokens and proposes the UniCodec, a universal speech token learning that encapsulates all semantics of speech, including linguistic and paralinguistic information, into a compact and semantically-disentangled unified token. Such a unified token can not only benefit speech language models in understanding with paralinguistic hints but also help speech generation with high-quality output. A low-bitrate neural codec is leveraged to learn such disentangled discrete representations at global and local scales, with knowledge distilled from self-supervised learned features. Extensive evaluations on multilingual datasets demonstrate its effectiveness in generating natural, expressive and long-term consistent output quality with paralinguistic attributes well preserved in several speech processing tasks.

cs.SD↗

arXiv2025

Latent-Domain Predictive Neural Speech Coding

Neural audio/speech coding has recently demonstrated its capability to deliver high quality at much lower bitrates than traditional methods. However, existing neural audio/speech codecs employ either acoustic features or learned blind features with a convolutional neural network for encoding, by which there are still temporal redundancies within encoded features. This paper introduces latent-domain predictive coding into the VQ-VAE framework to fully remove such redundancies and proposes the TF-Codec for low-latency neural speech coding in an end-to-end manner. Specifically, the extracted features are encoded conditioned on a prediction from past quantized latent frames so that temporal correlations are further removed. Moreover, we introduce a learnable compression on the time-frequency input to adaptively adjust the attention paid to main frequencies and details at different bitrates. A differentiable vector quantization scheme based on distance-to-soft mapping and Gumbel-Softmax is proposed to better model the latent distributions with rate constraint. Subjective results on multilingual speech datasets show that, with low latency, the proposed TF-Codec at 1 kbps achieves significantly better quality than Opus at 9 kbps, and TF-Codec at 3 kbps outperforms both EVS at 9.6 kbps and Opus at 12 kbps. Numerous studies are conducted to demonstrate the effectiveness of these techniques. Code and models are available at https://github.com/microsoft/TF-Codec.

cs.SD↗

arXiv2025

Bitrate-Controlled Diffusion for Disentangling Motion and Content in Video

We propose a novel and general framework to disentangle video data into its dynamic motion and static content components. Our proposed method is a self-supervised pipeline with less assumptions and inductive biases than previous works: it utilizes a transformer-based architecture to jointly generate flexible implicit features for frame-wise motion and clip-wise content, and incorporates a low-bitrate vector quantization as an information bottleneck to promote disentanglement and form a meaningful discrete motion space. The bitrate-controlled latent motion and content are used as conditional inputs to a denoising diffusion model to facilitate self-supervised representation learning. We validate our disentangled representation learning framework on real-world talking head videos with motion transfer and auto-regressive motion generation tasks. Furthermore, we also show that our method can generalize to other types of video data, such as pixel sprites of 2D cartoon characters. Our work presents a new perspective on self-supervised learning of disentangled video representations, contributing to the broader field of video analysis and generation.

cs.CV↗

arXiv2024

Convert and Speak: Zero-shot Accent Conversion with Minimum Supervision

Low resource of parallel data is the key challenge of accent conversion(AC) problem in which both the pronunciation units and prosody pattern need to be converted. We propose a two-stage generative framework "convert-and-speak" in which the conversion is only operated on the semantic token level and the speech is synthesized conditioned on the converted semantic token with a speech generative model in target accent domain. The decoupling design enables the "speaking" module to use massive amount of target accent speech and relieves the parallel data required for the "conversion" module. Conversion with the bridge of semantic token also relieves the requirement for the data with text transcriptions and unlocks the usage of language pre-training technology to further efficiently reduce the need of parallel accent speech data. To reduce the complexity and latency of "speaking", a single-stage AR generative model is designed to achieve good quality as well as lower computation cost. Experiments on Indian-English to general American-English conversion show that the proposed framework achieves state-of-the-art performance in accent similarity, speech quality, and speaker maintenance with only 15 minutes of weakly parallel data which is not constrained to the same speaker. Extensive experimentation with diverse accent types suggests that this framework possesses a high degree of adaptability, making it readily scalable to accommodate other accents with low-resource data. Audio samples are available at https://www.microsoft.com/en-us/research/project/convert-and-speak-zero-shot-accent-conversion-with-minimumsupervision/.

cs.SD↗

arXiv2024

QDFormer: Towards Robust Audiovisual Segmentation in Complex Environments with Quantization-based Semantic Decomposition

Audiovisual segmentation (AVS) is a challenging task that aims to segment visual objects in videos according to their associated acoustic cues. With multiple sound sources and background disturbances involved, establishing robust correspondences between audio and visual contents poses unique challenges due to (1) complex entanglement across sound sources and (2) frequent changes in the occurrence of distinct sound events. Assuming sound events occur independently, the multi-source semantic space can be represented as the Cartesian product of single-source sub-spaces. We are motivated to decompose the multi-source audio semantics into single-source semantics for more effective interactions with visual content. We propose a semantic decomposition method based on product quantization, where the multi-source semantics can be decomposed and represented by several disentangled and noise-suppressed single-source semantics. Furthermore, we introduce a global-to-local quantization mechanism, which distills knowledge from stable global (clip-level) features into local (frame-level) ones, to handle frequent changes in audio semantics. Extensive experiments demonstrate that our semantically decomposed audio representation significantly improves AVS performance, e.g., +21.2% mIoU on the challenging AVS-Semantic benchmark with ResNet50 backbone. https://github.com/lxa9867/QSD.

cs.CV↗

arXiv2024

Masked Audio Modeling with CLAP and Multi-Objective Learning

Most existing masked audio modeling (MAM) methods learn audio representations by masking and reconstructing local spectrogram patches. However, the reconstruction loss mainly accounts for the signal-level quality of the reconstructed spectrogram and is still limited in extracting high-level audio semantics. In this paper, we propose to enhance the semantic modeling of MAM by distilling cross-modality knowledge from contrastive language-audio pretraining (CLAP) representations for both masked and unmasked regions (MAM-CLAP) and leveraging a multi-objective learning strategy with a supervised classification branch (SupMAM), thereby providing more semantic knowledge for MAM and enabling it to effectively learn global features from labels. Experiments show that our methods significantly improve the performance on multiple downstream tasks. Furthermore, by combining our MAM-CLAP with SupMAM, we can achieve new state-of-the-art results on various audio and speech classification tasks, exceeding previous self-supervised learning and supervised pretraining methods.

cs.SD↗

arXiv2024

Low-latency Speech Enhancement via Speech Token Generation

Existing deep learning based speech enhancement mainly employ a data-driven approach, which leverage large amounts of data with a variety of noise types to achieve noise removal from noisy signal. However, the high dependence on the data limits its generalization on the unseen complex noises in real-life environment. In this paper, we focus on the low-latency scenario and regard speech enhancement as a speech generation problem conditioned on the noisy signal, where we generate clean speech instead of identifying and removing noises. Specifically, we propose a conditional generative framework for speech enhancement, which models clean speech by acoustic codes of a neural speech codec and generates the speech codes conditioned on past noisy frames in an auto-regressive way. Moreover, we propose an explicit-alignment approach to align noisy frames with the generated speech tokens to improve the robustness and scalability to different input lengths. Different from other methods that leverage multiple stages to generate speech codes, we leverage a single-stage speech generation approach based on the TF-Codec neural codec to achieve high speech quality with low latency. Extensive results on both synthetic and real-recorded test set show its superiority over data-driven approaches in terms of noise robustness and temporal speech coherence.

cs.SD↗

arXiv2023

ABC-KD: Attention-Based-Compression Knowledge Distillation for Deep Learning-Based Noise Suppression

Noise suppression (NS) models have been widely applied to enhance speech quality. Recently, Deep Learning-Based NS, which we denote as Deep Noise Suppression (DNS), became the mainstream NS method due to its excelling performance over traditional ones. However, DNS models face 2 major challenges for supporting the real-world applications. First, high-performing DNS models are usually large in size, causing deployment difficulties. Second, DNS models require extensive training data, including noisy audios as inputs and clean audios as labels. It is often difficult to obtain clean labels for training DNS models. We propose the use of knowledge distillation (KD) to resolve both challenges. Our study serves 2 main purposes. To begin with, we are among the first to comprehensively investigate mainstream KD techniques on DNS models to resolve the two challenges. Furthermore, we propose a novel Attention-Based-Compression KD method that outperforms all investigated mainstream KD frameworks on DNS task.

eess.AS↗

arXiv2023

DasFormer: Deep Alternating Spectrogram Transformer for Multi/Single-Channel Speech Separation

For the task of speech separation, previous study usually treats multi-channel and single-channel scenarios as two research tracks with specialized solutions developed respectively. Instead, we propose a simple and unified architecture - DasFormer (Deep alternating spectrogram transFormer) to handle both of them in the challenging reverberant environments. Unlike frame-wise sequence modeling, each TF-bin in the spectrogram is assigned with an embedding encoding spectral and spatial information. With such input, DasFormer is then formed by multiple repetition of simple blocks each of which integrates 1) two multi-head self-attention (MHSA) modules alternately processing within each frequency bin & temporal frame of the spectrogram 2) MBConv before each MHSA for modeling local features on the spectrogram. Experiments show that DasFormer has a powerful ability to model the time-frequency representation, whose performance far exceeds the current SOTA models in multi-channel speech separation, and also achieves single-channel SOTA in the more challenging yet realistic reverberation scenario.

cs.SD↗

arXiv2023

Contrast-PLC: Contrastive Learning for Packet Loss Concealment

Packet loss concealment (PLC) is challenging in concealing missing contents both plausibly and naturally when there are only limited available context to use. Recently deep-learning based PLC algorithms have demonstrated their superiority over traditional counterparts; but their concealment ability is still mostly limited to a maximum of 120ms loss. Even with strong GAN-based generative models, it is still very challenging to predict long burst losses that could happen within/in-between phonemes. In this paper, we propose to use contrastive learning to learn a loss-robust semantic representation for PLC. A hybrid neural PLC architecture combining the semantic prediction and GAN-based generative model is designed to verify its effectiveness. Results on the blind test set of Interspeech2022 PLC Challenge show its superiority over commonly used UNet-style framework and the one without contrastive learning, especially for the longer burst loss at (120, 220] ms.

cs.SD↗

arXiv2023

Disentangled Feature Learning for Real-Time Neural Speech Coding

Recently end-to-end neural audio/speech coding has shown its great potential to outperform traditional signal analysis based audio codecs. This is mostly achieved by following the VQ-VAE paradigm where blind features are learned, vector-quantized and coded. In this paper, instead of blind end-to-end learning, we propose to learn disentangled features for real-time neural speech coding. Specifically, more global-like speaker identity and local content features are learned with disentanglement to represent speech. Such a compact feature decomposition not only achieves better coding efficiency by exploiting bit allocation among different features but also provides the flexibility to do audio editing in embedding space, such as voice conversion in real-time communications. Both subjective and objective results demonstrate its coding efficiency and we find that the learned disentangled features show comparable performance on any-to-any voice conversion with modern self-supervised speech representation learning models with far less parameters and low latency, showing the potential of our neural coding framework.

cs.SD↗

arXiv2023

Time-Variance Aware Real-Time Speech Enhancement

Time-variant factors often occur in real-world full-duplex communication applications. Some of them are caused by the complex environment such as non-stationary environmental noises and varying acoustic path while some are caused by the communication system such as the dynamic delay between the far-end and near-end signals. Current end-to-end deep neural network (DNN) based methods usually model the time-variant components implicitly and can hardly handle the unpredictable time-variance in real-time speech enhancement. To explicitly capture the time-variant components, we propose a dynamic kernel generation (DKG) module that can be introduced as a learnable plug-in to a DNN-based end-to-end pipeline. Specifically, the DKG module generates a convolutional kernel regarding to each input audio frame, so that the DNN model is able to dynamically adjust its weights according to the input signal during inference. Experimental results verify that DKG module improves the performance of the model under time-variant scenarios, in the joint acoustic echo cancellation (AEC) and deep noise suppression (DNS) tasks.

eess.AS↗

arXiv2023

Improving Speech Enhancement via Event-based Query

Existing deep learning based speech enhancement (SE) methods either use blind end-to-end training or explicitly incorporate speaker embedding or phonetic information into the SE network to enhance speech quality. In this paper, we perceive speech and noises as different types of sound events and propose an event-based query method for SE. Specifically, representative speech embeddings that can discriminate speech with noises are first pre-trained with the sound event detection (SED) task. The embeddings are then clustered into fixed golden speech queries to assist the SE network to enhance the speech from noisy audio. The golden speech queries can be obtained offline and generalizable to different SE datasets and networks. Therefore, little extra complexity is introduced and no enrollment is needed for each speaker. Experimental results show that the proposed method yields significant gains compared with baselines and the golden queries are well generalized to different datasets.

cs.SD↗

arXiv2023

Real-time speech enhancement with dynamic attention span

For real-time speech enhancement (SE) including noise suppression, dereverberation and acoustic echo cancellation, the time-variance of the audio signals becomes a severe challenge. The causality and memory usage limit that only the historical information can be used for the system to capture the time-variant characteristics. We propose to adaptively change the receptive field according to the input signal in deep neural network based SE model. Specifically, in an encoder-decoder framework, a dynamic attention span mechanism is introduced to all the attention modules for controlling the size of historical content used for processing the current frame. Experimental results verify that this dynamic mechanism can better track time-variant factors and capture speech-related characteristics, benefiting to both interference removing and speech quality retaining.

eess.AS↗