arXiv subjects

Zhiyue Wu

Publications and source records attributed to Zhiyue Wu.

14 recordsLinked to original sources

arXiv2026

StepAudio 3 Realtime Technical Report

Realtime spoken interaction demands deep reasoning, prompt responses, and fluid turn-taking. We present StepAudio 3 Realtime, an audio-language foundation model organized around a continuous listen-converse-think-act loop. Deep Perception captures rich acoustic cues to interpret user intent, while Seamless Duplex models synchronized audio streams to handle pauses, backchannels, and interruptions naturally. Crucially, we resolve the tension between deep deliberation and latency via Think-While-Speaking, executing private reasoning in parallel with spoken delivery. In reasoning mode, StepAudio 3 reaches a 73.0 macro average on StepAudioChat. With Think-While-Speaking, it achieves dialogue and reasoning performance comparable to dedicated reasoning models while speaking in real time. Furthermore, an integrated Voice Agent handles asynchronous tool execution without disrupting the dialogue flow. StepAudio 3 Realtime achieves top-tier performance across key dimensions: an exceptional 90.6 on the MMSU benchmark, 98.9 Overall on the Artificial Analysis Full-Duplex Bench, and a 56.0% macro task-success rate on $τ$-Voice.

cs.SD↗

arXiv2026

Structure-Fair Quantum Circuit Complexity: An Auditable Information-Theoretic Lower Bound

Quantum circuit complexity is often used to characterize the physical cost of state preparation, but its physical meaning depends on the reference and counting rules; the entropy-removal costs of operations such as reset may be left out of resource accounting. We propose the principle of structural fairness and develop the Reference-Contingent Complexity (RCC) framework, jointly specifying the reference, generation capabilities, and atomic costs. We construct a model family that can approximate arbitrary finite-dimensional pure and mixed states. Within an admissible model fixed in advance, we prove a rigorous lower bound on universal optimal quantum circuit complexity. The target state's smooth one-shot information gap relative to the unbiased structured vacuum (the maximum-entropy state on the reference support) has an entropy-spectrum structure. Calibrated by the atomic control bandwidth and with finite-description corrections included, this gap sets a common cost floor for every admissible successful path. Predeclared final-state measurements and their finite-sample statistics thus yield independently verifiable one-sided complexity lower-bound certificates without reconstructing the generation history. Finally, exact structural allocation relations under changes of observation window and reference motivate a conjecture on the reference covariance of entropy and complexity: a reference can shift the complexity zero point, but cannot remove the burden of generating structure at no cost.

quant-ph↗

arXiv2026

DuoTok: Source-Aware Dual-Track Music Tokenization for Vocal-Accompaniment Generation

Multi-track music generation requires tokens that preserve acoustic fidelity, support sequence modeling, and maintain cross-track structure. Reconstruction-oriented codecs retain acoustic detail but are difficult to model, while semantic tokenizers may sacrifice fidelity or cross-track alignment. We present DuoTok, a source-aware dual-track music tokenizer for vocal-accompaniment generation based on staged disentanglement. DuoTok first learns a semantic audio representation through self-supervised pretraining, then shapes source-aware structure using feature replacement noise and multi-task supervision: spectral reconstruction, music source separation regularization, and an ASR head for lyric alignment. It freezes the encoder and learns hard-routed codebooks for vocals and accompaniment, while a diffusion decoder restores fine acoustic detail from discrete tokens. Across public benchmarks, DuoTok achieves a favorable predictability-fidelity trade-off at ultra-low bitrate. Under held-constant dual-track language modeling, it improves both unconditional vocal-accompaniment modeling and vocal-conditioned accompaniment prediction. Controlled diagnostics show larger predictability costs under cross-track corruption and greater gains from longer temporal context, supporting stronger use of cross-track and temporal structure rather than merely easier local prediction. DuoTok also maintains competitive reconstruction quality and preserves control-relevant musical attributes in its discrete space. These results suggest that tokenizer design is a core modeling problem for multi-track music generation, beyond compression alone.

cs.SD↗

arXiv2026

StepAudio 3 Gen Technical Report

We introduce StepAudio 3 Gen, a general-purpose audio generation model that supports zero-shot text-to-speech (TTS), voice design, vocal generation, sound effects, music, vibe speech, and mixtures of multiple audio types within a unified framework. At its core, StepAudio 3 Gen is a discrete autoregressive generator that models audio directly over residual vector quantization (RVQ) tokens, departing from the diffusion Transformer-based continuous generation paradigm prevalent in recent general audio models. Its StepAudio Tokenizer represents general audio at 12.5 Hz in a shared $16 \times 2048$ residual code space, jointly quantizing semantic and waveform-level acoustic features so that each code layer preserves both types of information. For generation, the backbone predicts the first codebook along the time axis using autoregressive modeling, while a lightweight causal Transformer completes the remaining fifteen codebooks along the codebook axis. Our study further identifies three key design principles: (1) interference-aware progressive pretraining for acquiring audio capabilities while preserving the textual abilities of the large language model, (2) RVQ Adaptor for effectively incorporating multi-codebook acoustic representations, and (3) discrete autoregressive modeling over a shared representation across general audio domains. With progressive pretraining, multi-task instruction training, and supervised fine-tuning, StepAudio 3 Gen achieves state-of-the-art performance on both TTS and voice design, while retaining strong generation capabilities across speech, vocals, sound effects, and music. Audio samples are available at https://stepaudiollm.github.io/step-audio-3-gen/.

cs.SD↗

arXiv2026

StepAudio 3 Music Technical Report

We introduce StepAudio 3 Music, a large-scale, long-form music generation model that supports explicit musical planning and open-domain text-controlled generation. The StepAudio Music Tokenizer represents audio as a 50-Hz stream from a 65536-entry single codebook, using semantically informed self-supervised and multi-task training to preserve musical structure and reconstruction-relevant information. A flow-matching diffusion Transformer (DiT) predicts continuous StepAudio VAE latents, which our VAE decoder converts into 48-kHz audio. This discrete-continuous design is guided by comparisons of single-codebook VQ, Semantic and Acoustic RVQ, and different DiT configurations. For explicit planning, a Mixture-of-Experts autoregressive model uses ABC notation to produce an intermediate arrangement plan (ABC-CoT) before predicting music tokens, making harmony, rhythm, and melodic structure part of the generation context. A progressive training curriculum and supervised fine-tuning support song and instrumental generation, accompaniment generation from dry vocals, and cover-song synthesis for up to 5 minutes and 30 seconds. With reinforcement learning via direct preference optimization (DPO), the final model achieves the highest AudioBox Content Enjoyment, Content Usefulness, and Production Quality scores and the highest MuQ-MuLan similarity among the evaluated systems, with competitive SongBench results. On the preliminary Artificial Analysis Music Arena Vocals leaderboard, it obtains a Quality Elo of 1105, behind only Suno V5.5 and Mureka and ahead of Suno V5, MiniMax models, and other systems. Audio demonstrations are available at https://stepaudiollm.github.io/step-audio-3-music.

eess.AS↗

arXiv2026

StepAudio 2.5 Technical Report

Unified audio-language modeling has emerged as a prominent trend in modern speech systems, promising to bring the reasoning capabilities of large language models to auditory tasks. However, existing unified foundations often struggle to match the depth of specialized systems across automatic speech recognition (ASR), text-to-speech synthesis (TTS), and realtime spoken interaction. Bridging this gap remains an open challenge. This report presents StepAudio 2.5, a unified audio-language foundation model that matches or exceeds specialized systems across all three capabilities. Rather than treating these tasks as architecturally distinct, we operate on the premise that once text and audio share a multimodal representational space, task specialization becomes a matter of operational regimes: data construction, optimization targets, and decoding constraints. Guided by this insight, we advance the post-training paradigm from standard supervised learning to task-tailored Reinforcement Learning from Human Feedback (RLHF), using it as the primary mechanism to define complex optimization targets. We leverage this RLHF-centric alignment, alongside specialized decoding, to shape a shared backbone into three distinct operational modes. Concretely, the ASR branch advances transcription efficiency via verifiable multi-token decoding; the TTS branch achieves controllable, expressive synthesis through preference-based RLHF and context-rich supervision; and the Realtime branch realizes low-latency, persona-consistent dialogue via generative reward modeling within an RLHF framework. On standard benchmarks, StepAudio 2.5 achieves state-of-the-art results across ASR, TTS, and Realtime, demonstrating that a singular audio-language foundation can successfully internalize the distinct deployment objectives of speech understanding, generation, and live interaction.

eess.AS↗

arXiv2025

A Heuristic Study of Temperature: Quantum Circuitry in Thermal Systems

The singularities prevalent in classical thermodynamics largely stem from the "postulate of equal a priori probabilities" neglecting the physical constraints imposed by computational complexity. This paper introduces Complexity Window Thermodynamics (CWT), a framework that characterizes the observer's "ignorance" via a finite complexity budget, thereby naturally smoothing out singular behaviors associated with phase transitions and negative temperatures within this window. We derive a generalized First Law of Thermodynamics driven by a complexity generation potential, which incorporates "information processing work," and demonstrate a universal action-time bound constraining the growth of complexity. CWT not only offers a unified perspective on critical phenomena in condensed matter and the black hole information problem but also suggests that the total generatable complexity of the universe is comparable in order of magnitude to its holographic entropy. Thus, it paves a new pathway for a resource-theoretic unification of thermodynamics, quantum computation, and gravity.

cond-mat.stat-mech↗

arXiv2025

Back to Ear: Perceptually Driven High Fidelity Music Reconstruction

Variational Autoencoders (VAEs) are essential for large-scale audio tasks like diffusion-based generation. However, existing open-source models often neglect auditory perceptual aspects during training, leading to weaknesses in phase accuracy and stereophonic spatial representation. To address these challenges, we propose εar-VAE, an open-source music signal reconstruction model that rethinks and optimizes the VAE training paradigm. Our contributions are threefold: (i) A K-weighting perceptual filter applied prior to loss calculation to align the objective with auditory perception. (ii) Two novel phase losses: a Correlation Loss for stereo coherence, and a Phase Loss using its derivatives--Instantaneous Frequency and Group Delay--for precision. (iii) A new spectral supervision paradigm where magnitude is supervised by all four Mid/Side/Left/Right components, while phase is supervised only by the LR components. Experiments show εar-VAE at 44.1kHz substantially outperforms leading open-source models across diverse metrics, showing particular strength in reconstructing high-frequency harmonics and the spatial characteristics.

cs.SD↗

arXiv2025

AudioCodecBench: A Comprehensive Benchmark for Audio Codec Evaluation

Multimodal Large Language Models (MLLMs) have been widely applied in speech and music. This tendency has led to a focus on audio tokenization for Large Models (LMs). Unlike semantic-only text tokens, audio tokens must both capture global semantic content and preserve fine-grained acoustic details. Moreover, they provide a discrete method for speech and music that can be effectively integrated into MLLMs. However, existing research is unsuitable in the definitions of semantic tokens and acoustic tokens. In addition, the evaluation of different codecs typically concentrates on specific domains or tasks, such as reconstruction or Automatic Speech Recognition (ASR) task, which prevents fair and comprehensive comparisons. To address these problems, this paper provides suitable definitions for semantic and acoustic tokens and introduces a systematic evaluation framework. This framework allows for a comprehensive assessment of codecs' capabilities which evaluate across four dimensions: audio reconstruction metric, codebook index (ID) stability, decoder-only transformer perplexity, and performance on downstream probe tasks. Our results show the correctness of the provided suitable definitions and the correlation among reconstruction metrics, codebook ID stability, downstream probe tasks and perplexity.

cs.SD↗

arXiv2025

The Cost of Nonlocality: A Dynamical Performance Equation of Energy-Entanglement-Complexity

This work aims to quantify the physical cost of generating non-local entanglement in systems governed by local interactions. By unifying the quantum speed limit and Lieb-Robinson bounds, we establish an "energy-entanglement performance equation." This framework connects theoretical computational complexity with experimental observables by introducing a measurable proxy for complexity, thereby revealing a performance trade-off among the "energy variance-entanglement product," the strength of local interactions, and dynamical efficiency. Our work not only defines a "performance frontier"-constrained by theoretical bounds and amenable to experimental benchmarking-but also provides a novel diagnostic tool for identifying the performance bottlenecks of a process.

quant-ph↗

arXiv2025

The Exploratory Study on the Relationship Between the Failure of Distance Metrics in High-Dimensional Space and Emergent Phenomena

This paper presents a unified framework, integrating information theory and statistical mechanics, to connect metric failure in high-dimensional data with emergence in complex systems. We propose the "Information Dilution Theorem," demonstrating that as dimensionality ($d$) increases, the mutual information efficiency between geometric metrics (e.g., Euclidean distance) and system states decays approximately as $O(1/d)$. This decay arises from the mismatch between linearly growing system entropy and sublinearly growing metric entropy, explaining the mechanism behind distance concentration. Building on this, we introduce information structural complexity ($C(S)$) based on the mutual information matrix spectrum and interaction encoding capacity ($C'$) derived from information bottleneck theory. The "Emergence Critical Theorem" states that when $C(S)$ exceeds $C'$, new global features inevitably emerge, satisfying a predefined mutual information threshold. This provides an operational criterion for self-organization and phase transitions. We discuss potential applications in physics, biology, and deep learning, suggesting potential directions like MI-based manifold learning (UMAP+) and offering a quantitative foundation for analyzing emergence across disciplines.

cs.IT↗

arXiv2024

The Information Theory of Self-Organization Phenomena in Thermal Systems

This paper revisits Brownian motion from the perspective of Information Theory, aiming to explore the connections between Information Theory, Thermodynamics, and Complex Science. First, we propose a single-particle discrete Brownian motion model (SPBM). Within the framework of the maximum entropy principle and Bayesian inference, we demonstrate the equivalence of prior information and constraint conditions, revealing the relationship between local randomness and global probability distribution. By analyzing particle motion, we find that local constraints and randomness can lead to global probability distributions, thereby reflecting the interplay between local and global dynamics in the process of information transfer. Next, we extend our research to multi-particle systems, introducing the concepts of "Energy as Encoding" and "Information Temperature" to clarify how energy distribution determines information structure. We explore how energy, as not only a fundamental physical quantity in physical systems but also an inherently informational one, directly dictates the prior probability distribution of system states, thus serving as a form of information encoding. Based on this, we introduce the concept of "Equilibrium Flow" to explain the self-organizing behavior of systems under energy constraints and Negative Information Temperature. By proving three theorems regarding Equilibrium Flow systems, we reveal the criticality of Self-Organization, energy-information conversion efficiency, and the characteristic that event occurrence probabilities follow the Fermi-Dirac distribution. Through theoretical analysis and theorem proofs, we offer new perspectives for understanding the dynamics of Complex Systems, enriching the theoretical framework of Information Theory, Thermodynamics, and Complex Science, and providing a new theoretical basis for further research in related fields.

cond-mat.stat-mech↗

arXiv2024

Can LLMs "Reason" in Music? An Evaluation of LLMs' Capability of Music Understanding and Generation

Symbolic Music, akin to language, can be encoded in discrete symbols. Recent research has extended the application of large language models (LLMs) such as GPT-4 and Llama2 to the symbolic music domain including understanding and generation. Yet scant research explores the details of how these LLMs perform on advanced music understanding and conditioned generation, especially from the multi-step reasoning perspective, which is a critical aspect in the conditioned, editable, and interactive human-computer co-creation process. This study conducts a thorough investigation of LLMs' capability and limitations in symbolic music processing. We identify that current LLMs exhibit poor performance in song-level multi-step music reasoning, and typically fail to leverage learned music knowledge when addressing complex musical tasks. An analysis of LLMs' responses highlights distinctly their pros and cons. Our findings suggest achieving advanced musical capability is not intrinsically obtained by LLMs, and future research should focus more on bridging the gap between music knowledge and reasoning, to improve the co-creation experience for musicians.

cs.SD↗

arXiv2021

DRAM Failure Prediction in AIOps: Empirical Evaluation, Challenges and Opportunities

DRAM failure prediction is a vital task in AIOps, which is crucial to maintain the reliability and sustainable service of large-scale data centers. However, limited work has been done on DRAM failure prediction mainly due to the lack of public available datasets. This paper presents a comprehensive empirical evaluation of diverse machine learning techniques for DRAM failure prediction using a large-scale multi-source dataset, including more than three millions of records of kernel, address, and mcelog data, provided by Alibaba Cloud through PAKDD 2021 competition. Particularly, we first formulate the problem as a multi-class classification task and exhaustively evaluate seven popular/state-of-the-art classifiers on both the individual and multiple data sources. We then formulate the problem as an unsupervised anomaly detection task and evaluate three state-of-the-art anomaly detectors. Further, based on the empirical results and our experience of attending this competition, we discuss major challenges and present future research opportunities in this task.

cs.LG↗