[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-dynamic-band-splitting-audio-separation":3,"topics-all":38,"news-related-55663822-d49d-4c7e-a7c7-ba9770657c25":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"55663822-d49d-4c7e-a7c7-ba9770657c25","清华DBS动态分频:音源分离SI-SDR提升29%","音源分离的 band-split 架构用固定边界切频段,不随内容变。清华团队提出 DBS 让边界按每条混音自适应:边界分数预测、硬 Top-K、内容加权池化,band-token 预算不变。DnR v2 上 DBS-BSRNN 平均 SI-SDR 从 7.03 升到 9.07 dB,相对提升 29.0%。","频谱切分是音源分离模型的经典前置步骤:band-split 架构把频谱手动切成预设频段,每个频段编码成一个 token。问题是边界对所有输入都一样——而一段混音里信息密度高的位置是随内容变的,固定边界要么把有区分度的成分切进同一频段糊掉,要么在低信息区浪费 token。\n\n清华大学 Bingzhe Zhang、Kai Li、Xiaolin Hu 团队在论文《Learning Where to Split: Dynamic Band Splitting for Audio Source Separation》里给出的答案是:边界本身可学习、逐输入自适应,而 token 预算封顶不变。\n\n## DBS 怎么让边界动起来\n\nDynamic Band Splitting(DBS)是端到端框架,在固定 band-token 预算下让边界位置与频段宽度随每条混音自适应,流程三步:\n\n1. **边界预测**:时序卷积与逐频率 bin 幅度统计喂给 BiLSTM,为每个候选切分点输出边界概率;\n2. **硬 Top-K**:每个粗区域内选分数最高的切分点,形成恰好规定数量的连续频段,token 数恒定;\n3. **内容加权池化**:把频段信息压成紧凑 band token。\n\n分离完成后,token 被平滑扩展回原始 STFT 网格、沿频率精修,再用频率共享解码头输出复数掩码。\n\n## 数字长什么样\n\n论文在 DnR v2、MUSDB18-HQ(音乐)、EchoSet(语音)三个基准上对比 DBS 与对应静态分频模型,关键结果:\n\n- 相对文献已报告的 BSRNN 成绩,DBS-BSRNN 在 DnR v2 上把平均 SI-SDR 从 7.03 dB 提到 9.07 dB,相对提升 29.0%;\n- 同为 128 band-token 预算,静态 BSRNN 对照 7.69 dB,DBS 版 9.07 dB;TIGER 骨干从 8.74 提到 10.13 dB(一作在 HF 论文页评论区补充);\n- BSRNN、TIGER、BS-RoFormer、MelRoFormer 四种骨干全部获益。\n\n消融有两个值得细看的结论。其一,粗区域从 4 个砍到 2 个,平均 SI-SDR 从 9.07 掉到 8.32 dB,砍到 1 个掉到 6.30 dB——预算没变,边界全挤向低频。其二,把训练好的 DBS 在推理时换回固定 BSRNN 布局(不重训)得 8.62 dB,与单独训练的固定布局 8.67 dB 只差 0.05;纯边界摆放贡献 +0.45 dB,相对均匀切分 +1.27 dB。\n\n## 为什么值得看\n\n论文的核心姿态是把音频频谱当成 token 预算分配问题:预算封顶,分配随内容动——与 MoE 按需激活、动态深度推理在哲学上同构,资源总量不变,搬到信息密度高的地方。对音源分离领域,这说明 band-split 沿用多年的「手动定边界」默认做法确有可学习的余地。\n\n工程开放度完整:代码开源于 GitHub(JusperLee\u002FDynamic-Band-Splitting),DBS-BSRNN-DnR、DBS-TIGER-DnR 两个预训练模型上了 Hugging Face,另有可视化边界的 HF 试听 Demo。\n\n一句提醒:29.0% 是论文口径,分母是文献已报告的 BSRNN 成绩;换自训静态对照时绝对数字不同(7.69 vs 7.03),引用时看清分母。对语音前端、影视音频后期、音乐检索团队,这是可直接跑 demo 的开源工件。下一个值得追问的是:这套「学切分位置」的机制能否搬到其他按网格切 token 的模态,比如视频的时空块切分。\n\n参考:arXiv:2610.14993;项目页 cslikai.cn\u002FDynamic-Band-Splitting","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.14993","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"7ac06d8e-b074-4147-abfc-ffaa4c6b8744","ai-efficiency",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",{"id":19,"name":20,"slug":20,"description":14,"color":14},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":22,"name":23,"slug":23,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"7eae7c98-acb2-4516-9c0d-08ea339df26d","en","DBS Learns Where to Split: Audio Separation Gains 29% SI-SDR","Tsinghua DBS learns adaptive frequency boundaries for audio separation under a fixed band-token budget; DnR v2 SI-SDR rises 29.0% over BSRNN.","Spectrum partitioning is a classic front-end step for audio source separation models: band-split architectures manually carve the spectrum into predefined frequency bands, encode each band as a token, and hand them to the separation network. The catch is that these boundaries are identical for every input, while \"where the information density sits\" in a mixture changes with content. Fixed boundaries either merge discriminative spectral components into one band, or spend tokens where little resolution is needed.\n\nA Tsinghua University team — Bingzhe Zhang, Kai Li, and Xiaolin Hu — answers differently in the paper \"Learning Where to Split: Dynamic Band Splitting for Audio Source Separation\": make the boundaries themselves learnable and input-adaptive, while the token budget stays capped.\n\n## How DBS moves the boundaries\n\nDynamic Band Splitting (DBS) is an end-to-end framework that adapts boundary locations and band widths to each mixture under a fixed band-token budget, in three steps:\n\n1. **Boundary prediction**: temporal convolutions plus per-bin magnitude statistics feed a BiLSTM, which outputs a boundary probability for every candidate transition; contextual dissimilarity and magnitude change drive the score.\n2. **Hard Top-K selection**: the highest-scoring transitions per coarse region are chosen, forming exactly the prescribed number of contiguous bands — the token count never moves.\n3. **Content-weighted pooling**: band contents are compressed into compact band tokens.\n\nAfter separation, the model smooths and expands these tokens back to the original STFT grid, refines them along frequency, and decodes complex masks with a frequency-shared head.\n\n## What the numbers say\n\nThe paper compares DBS against corresponding static-band models on DnR v2, MUSDB18-HQ (music separation), and EchoSet (speech separation):\n\n- Against the reported BSRNN result in the literature, DBS-BSRNN lifts average SI-SDR on DnR v2 from 7.03 dB to 9.07 dB, a relative gain of 29.0%.\n- Under the same 128 band-token budget, the static BSRNN control reaches 7.69 dB versus 9.07 dB for DBS-BSRNN; the TIGER backbone goes from 8.74 dB to 10.13 dB (numbers the first author added in the Hugging Face paper-page comment section).\n- All four backbones — BSRNN, TIGER, BS-RoFormer, and MelRoFormer — improve.\n\nTwo ablation findings deserve attention. First, cutting the number of coarse regions from four to two drops average SI-SDR from 9.07 dB to 8.32 dB; a single region falls to 6.30 dB — the token budget is unchanged, but boundaries crowd into low frequencies while the upper spectrum goes barely partitioned. Second, substituting the fixed BSRNN layout into a trained DBS model at inference (no retraining) reaches 8.62 dB, only 0.05 dB below the separately trained fixed-layout model at 8.67 dB; boundary placement alone accounts for +0.45 dB over that layout and +1.27 dB over a uniform split.\n\n## Why it matters\n\nThe core stance of this paper is treating the audio spectrum as a token-budget allocation problem: cap the budget, let the allocation follow the content. Philosophically this mirrors dynamic computation on the LLM side — mixture-of-experts routing on demand, dynamic-depth inference — keep total resources fixed, but move them to where the information density is. For the source-separation field, it shows that the long-standing \"hand-set boundaries\" default of band-split architectures indeed left learnable slack on the table.\n\nOpenness is complete on the engineering side: code is on GitHub (JusperLee\u002FDynamic-Band-Splitting), two pretrained checkpoints — DBS-BSRNN-DnR and DBS-TIGER-DnR — are on Hugging Face, plus an interactive HF demo where you can listen and watch predicted boundaries move. The paper itself runs 25 pages with 11 figures, with thorough ablations and boundary analyses.\n\nOne caution: the 29.0% figure is the paper's own relative gain against the reported BSRNN result; with a self-trained static control as baseline the absolute numbers differ (7.69 vs 7.03), so watch the denominator when citing. For teams working on speech front-ends, film audio post-production, or music information retrieval, this is an open artifact you can demo immediately. The next question worth asking: can this \"learn where to split\" mechanism transfer to other grid-tokenized modalities — say, spatiotemporal patching in video.\n\nReference: arXiv:2610.14993 (https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.14993); project page https:\u002F\u002Fcslikai.cn\u002FDynamic-Band-Splitting","dynamic-band-splitting-audio-separation","2026-10-21T15:02:40Z","2026-10-21T15:12:30.262917Z","2026-10-21T15:12:30.262928Z",true,"agent",51,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"6ab8e203-ac77-4653-9b17-d3f6a362d38b","VLX-Flow：把视频理解从「请求-响应」改造成「持续观察」的边缘 VLM","vlx-flow-edge-video-continuous","2026-06-26T22:08:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"a491d4db-2d5a-4723-b093-28d1512fd647","VLA-Δ:物理状态变化当监督信号,LIBERO 98.4","vla-delta-physical-state-transition","2026-10-20T19:07:27+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"9103452d-d036-4b0f-bee6-5eda4b7fbf66","Fastino 340M 分类模型:官方基准胜 4B 生成式","fastino-gliner2-5-decide-classifier","2026-09-26T13:09:43+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"d78ea8a8-bebb-4b94-b340-127eb2874a73","Ovis 全模态嵌入 3B:综合分领先 5.19","ovis-omni-embedding-3b","2026-09-23T21:08:45+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"12c67d52-17a2-4df5-8386-35d18ffd221a","JEPA-Anything:一套预测框架打通七个领域,湿实验也给了背书","jepa-anything-orthogonal-predictive-factorization","2026-09-19T23:10:37+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"550cee5e-18e8-4236-9304-7207ebc221a8","Agnes 3.0 Flash 开源:72 层仅 18 层带 KV 缓存,33B 单卡跑 262k 上下文","agnes-3-0-flash-preview-open-weights","2026-09-13T15:20:00+00:00"]