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。

2026-09-25
日期2026-09-25
已评分—
均分—
最高—

Daily Papers — 2026-09-25

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

1. A Comprehensive Study of Content Representations for Speech Synthesis

Authors: Diego Torres, Axel Roebel, Nicolas Obin

Categories: eess.AS, cs.LG, cs.SD | 5 pages, 1 figure Score: 5.95/10 (Obj:7 Id:5 Ind:7 Comp:8 Eff:6 Nov:5 Rep:5)

  • Strength: 去掉说话人通道的统一重合成框架把表征分成两个区间，HuBERT 的 EER 1.3% 对 VQ-ASR 64 码的 48.4%，且 decoder 放大三倍后次序不变。
  • Weakness: 公开权重的表征全部不晚于 2024 年，S3Tokenizer V1 50 Hz、Vevo content tokenizer、FocalCodec 50 Hz 均有公开权重却未评测，CosyVoice/Vevo 路线只用 460 h 小复现代替。
  • Full review: Claude Code 全文七公理审稿

Speech content representations are central to voice conversion, speech-to-speech translation, and multimodal language models, yet they are rarely compared under a common generative framework that directly measures what each representation contains. We address this by training a generative model conditioned solely on each representation and evaluating the generated audio along the content, speaker identity, and prosody axes. Across SSL features, supervised tokens, posteriorgrams, and neural audio codecs, we find two distinct regimes: representations that nearly reconstruct the original audio, and representations that effectively disentangle speaker identity. These results show that disentanglement depends not on supervision alone, but on the interaction between the training objective and the representation’s information capacity: supervised representations only disentangle speaker identity when their capacity is sufficiently constrained.

2. VoiceNet: Fine-Grained Voice Understanding Beyond Emotion at Scale

Authors: Christoph Schuhmann, Robert Kaczmarczyk, Gollam Rabby, Felix Friedrich, Maurice Kraus et al.

Categories: cs.SD, cs.AI, eess.AS | 33 pages, 6 figures, 8 tables. Christoph Schuhmann and Robert Kaczmarczyk contributed equally. Code and benchmark: https://github.com/LAION-AI/emolia-bench Score: 5.26/10 (Obj:6 Id:4 Ind:4 Comp:6 Eff:5 Nov:5 Rep:8)

  • Strength: 在 $3{,}944$ 条真实野外语音上做了三专家标注的 40 类情绪基准并全量开源，VoiceCLAP-Large 对底座的 held-out 领先 $+0.041$（CI 不跨零）。
  • Weakness: 挑选基准正样本的 EmpathicInsight-Voice 分类器没有作为基线报告，按论文比例粗算它的 affirmative 先验约 $0.716$，与最好模型的 $0.702$ 持平，循环构造的影响没有被量化。
  • Full review: Claude Code 全文七公理审稿

Expressive speech synthesis has outpaced expressive speech perception: systems now render fine-grained vocal performances that no public benchmark can score. Most benchmarks for this inverse problem stop at six to nine basic emotion categories, largely on acted speech. This paper introduces VoiceNet, a human-annotated representation-level benchmark for voice performance understanding on permissively-licensed in-the-wild speech. VoiceNet has two subsets: VoiceNet-Emo applies a 40-emotion taxonomy with three expert ratings per item, and VoiceNet-Ext, a preliminary subset, scores 57 talking-style attributes including speaking rate, vocal tension, breathiness, and register. The paper also releases Emolia, an emotion-annotated version of the Emilia corpus, with a curated rebalanced subset enriched by dense MOSS-Audio Thinking annotations. Two voice-text contrastive baselines train on this data: a 110M-parameter VoiceCLAP-Small for fast large-scale data filtering and a 7B VoiceCLAP-Large for state-of-the-art performance. Both outperform existing CLAP baselines, which sit near chance on VoiceNet-Emo. On VoiceNet-Emo, VoiceCLAP-Large aligns more closely with the aggregate expert consensus than individual experts agree with one another: a comparison against the majority label rather than evidence of surpassing human emotion perception. All systems evaluated here are voice-text embedding models: VoiceNet scores representation-level attribute recognition and retrieval, not end-to-end spoken-dialogue behaviour. Clustering and filtering uncurated speech corpora into subsets that span diverse talking styles and emotions remains an open challenge; VoiceCLAP embeddings offer a promising tool for this task. VoiceNet, Emolia, and VoiceCLAP are publicly available for research use.

3. TinyAudio: Compact and Efficient Text-to-Audio Generation for Low-Resource Deployment

Authors: Junxi Liu, Xiquan Li, Wenhao Guan, Yifan Duan, Zhikang Niu et al.

Categories: cs.SD, eess.AS Score: 4.84/10 (Obj:7 Id:4 Ind:4 Comp:8 Eff:5 Nov:4 Rep:3)

  • Strength: 在论文的单卡 FP32 设置下，$87$M 总参数、$0.48$ GB 峰值显存，比最省的基线少约 $8$ 倍显存，同时 AudioCaps FAD 维持在 $2.65$。
  • Weakness: 没有和任何同尺寸的公开紧凑模型（Stable Audio Open Small、AudioGAN、MeanAudio-S-Full）比较，而对大模型基线在 AudioCaps 各项指标上全部落后，FD 差约 $30\%$。
  • Full review: Claude Code 全文七公理审稿

Text-to-audio (TTA) generation has advanced rapidly in generation quality and instruction following. However, representative systems often require around a billion parameters, limiting deployment on resource-constrained devices. This paper introduces TinyAudio, a compact flow-matching-based TTA model for low-resource deployment. At its core, TinyAudio uses TA-DiT, a 35M single-stream flow-matching Transformer. TinyAudio also includes TA-CLAP, a 32M audio-aligned text encoder, and TA-VAE, whose 20M decoder reconstructs 44.1 kHz audio from compressed latents. TinyAudio has only 87M parameters in total, over 90% fewer than representative billion-parameter pipelines, and uses 0.48 GB peak GPU memory. TinyAudio achieves competitive generation quality on AudioCaps and TTA-Bench. We further introduce TinyAudio-MF, a MeanFlow-accelerated model that enables real-time generation with a four-core CPU quota. Our results demonstrate a practical quality-footprint trade-off for low-resource TTA deployment.