每日 arXiv

音频 · 语音 · 音乐 · 声学

每日新爬取的 arXiv 论文，按评分排序。中文摘要帮你快速判断是否值得深读。

2026-09-26
日期2026-09-26
已评分—
均分—
最高—

Daily Papers — 2026-09-26

3 papers on audio, speech, music, and acoustics.

1. SAGE: Semantic Audio Generative Encoder

Authors: Francesco Brigante, Luca Cerovaz, Davide Marincione, Giorgio Strano, Luca Zhou et al.

Categories: cs.SD, cs.LG, eess.AS | 18 pages, 6 figures, 11 tables. Code and weights: https://github.com/francescobrigante/SAGE. Project page: https://sage-music.pages.dev/ Score: 6.58/10 (Obj:7 Id:7 Ind:5 Comp:7 Eff:7 Nov:5 Rep:9)

  • Strength: 在论文的 FMA test 上，$105$M 的 SAGE 以与 Stable Audio Open 相同的 RTF（$0.0045$）把 FAD-MERT 做到 $0.113$，约为 $852$M 的 SAME-L（$0.210$）的一半，MUSHRA 与之持平（$81.6$ 对 $81.8$）。
  • Weakness: 语义优势来自 CLAP 蒸馏，去掉教师后探针从 $0.642$ 跌到 $0.495$，而论文没有在 SAGE latent 上训练任何下游生成模型，“语义 latent 更利于生成”的核心动机没有证据。
  • Full review: Claude Code 全文七公理审稿

Audio autoencoders compress waveforms into compact latent representations that serve as the interface between raw audio and downstream models. Current systems navigate a three-way trade-off between reconstruction quality, semantic structure of the latent space, and inference speed, typically favoring one or two of these at the expense of the others. This paper introduces SAGE, Semantic Audio Generative Encoder: a compact variational autoencoder, trained solely on publicly available music, that shapes its latent by distilling embeddings from a pretrained audio-text model. This 105M-parameter model runs at the inference cost of Stable Audio Open and reaches the listening-test quality of SAME-L, an autoencoder 8x larger and 4x slower, while surpassing both on objective perceptual and distributional metrics of reconstruction. Furthermore, it sets the state of the art on all nineteen probing tasks of latent semantics, in domain and out of domain. These results establish SAGE as a lightweight audio autoencoder that strikes the best balance of the three-way trade-off among those we evaluate, combining high reconstruction fidelity, state-of-the-art semantic structure, and fast inference.

2. Toward Human-Aligned Judgement of Speech Emotion Similarity

Authors: Yun-Shao Tsai, Yi-Cheng Lin, Chih-Kai Yang, Ho-Jung Cheng, Tsun-Yi Chang et al.

Categories: eess.AS, cs.SD | 5 pages Score: 6.26/10 (Obj:7 Id:5 Ind:8 Comp:8 Eff:6 Nov:6 Rep:5)

  • Strength: 在同一冻结 WavLM-SER 编码器下，仅把监督从 VAD 标签换成人类分级比较，Spearman 从 $0.4820$ 升到 $0.5297$，配对检验显著（Table 2）。
  • Weakness: LALM 基线只做了零样本提示且 SAME 记错，没有人类标注者一致性上限，也没有多 seed 方差，数据集和权重尚未放出。
  • Full review: Claude Code 全文七公理审稿

Evaluating emotion preservation in expressive speech generation involves assessing how closely generated speech matches a reference in emotion. Human listening tests assess this similarity, but their cost motivates automatic measures aligned with human judgments. To support the development and evaluation of such measures, we introduce SES-Bench, a speech emotion similarity benchmark built from human comparisons of two candidate utterances against a shared reference. These comparisons record which candidate listeners find emotionally closer to the reference and the strength of their preference. Using these annotations, we train SES-Judge to score emotion similarity between two utterances. SES-Judge significantly outperforms embedding cosine similarity and prompted large audio-language models in preference accuracy and correlation with human ratings that capture both preference direction and strength.

3. VoxMem: Benchmarking Multimodal Memory in Large Audio Language Models

Authors: Yang Xiao, Vidhyasaharan Sethu, Eun-Jung Holden, Ting Dang

Categories: eess.AS, cs.SD | working in process Score: 6.26/10 (Obj:7 Id:5 Ind:5 Comp:8 Eff:6 Nov:6 Rep:8)

  • Strength: 多会话口语记忆评测，四档历史固定题目做配对比较，在论文设置中 32K 时闭源模型语义记忆 $55.6\%$ 对副语言 $20.0\%$，差距远超噪声。
  • Weakness: 质检过滤、过滤验证与打分共用 Gemini-3.7-Flash 形成闭环，且缺 OpenAI 音频模型、Step-Audio 2 等强基线与 ASR 级联、人类参照，衰减速率排序在自家 Table 18 的重叠区间下得不到支持。
  • Full review: Claude Code 全文七公理审稿

Spoken conversational systems must recover information from prior interactions (i.e., memory), yet relevant information in speech extends beyond what was said to who said it, how it was spoken, and what was audible, information that exists only in the audio signal and cannot be recovered from a transcript. Beyond what to remember, memory also demands diverse operations: retrieving a single fact, integrating evidence across turns, tracking an evolving state. Real interactions further unfold across sessions, meaning information accumulates across distinct episodes rather than a single continuous recording. Existing benchmarks fall short on all three dimensions: they focus primarily on lexical content, adopt limited and ad hoc memory operations, and treat memory as a single-session problem. We argue that principled memory evaluation requires jointly characterizing the acoustic evidence to be retained and the operations applied to it, and introduce a taxonomy along these two axes. Building on this taxonomy, we present VoxMem: 3,196 evaluation instances over 34,743 spoken sessions (177 hours) crossing four acoustic evidence types (speech semantics, speaker identity, paralinguistic cues, environmental sound) with four memory operations (information extraction, multi-session reasoning, temporal tracking, and answer refusal), grounded in multi-session histories and stratified across context budgets from 8K to 64K tokens. Evaluating 15 LALMs, no model exceeds 40% at 32K. Models retain what was said far better than who said it, how, or what was audible, a gap that widens for complex operations, grows with history length, and manifests as qualitatively distinct failure modes across evidence types. VoxMem aims to provide a foundation to measure and drive progress on the full scope of spoken conversational memory.