Trae/agent 4 ch kjo - #112
Closed
lyk-bit wants to merge 13 commits into
Closed
Conversation
新增 ja-zh 翻译方向、日语 Whisper 识别和专用日译中提示词，补充前端选项、双语文档及回归测试。
- 修复 setup 创建 venv 时把解释器路径当作目标目录，导致生成 .venv\Scripts\python.exe 目录、后续 Assert-Python312 报找不到命令， 且二次运行无法自愈；合并重复分支并增加创建后断言 - 解释器探测改用 -PathType Leaf，并清理旧版遗留的同名目录， 使已损坏的安装可直接重跑恢复 - 改为按行重写 .env 密码哈希，避免 Argon2 哈希中的 $+ 与 $& 被 [regex]::Replace 当作替换标记破坏，同时收敛重复键 - 新增 stop-youdub.bat，并让 stop 从 release-manifest.json 读取 api_port Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Next 的 /api rewrite 会克隆请求体，默认上限 10MB，超出后静默截断并 提前发 EOF。后端按 Content-Length 死等剩余数据，30 秒 proxyTimeout 到点后连接重置，浏览器报 Failed to fetch，任务详情页也不会跳转。 - 设置 experimental.proxyClientMaxBodySize 为 8gb，高于后端 LOCAL_UPLOAD_MAX_BYTES 默认值，让后端保持权威限制并能正常返回 413 - 设置 experimental.proxyTimeout，30 秒默认值不足以传输 GB 级文件 - README 补充两个上限的耦合关系说明 实测：50MB 从 30 秒超时失败变为 0.43 秒 201；1.2GB 为 9.37 秒 201， 落盘字节数与源文件完全一致；超限时返回带文案的 413。 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Demucs 按整条音轨一次性分配 float32 张量（4 音轨 x 2 声道 x 4 字节 = 每采样点 32 字节）。htdemucs_ft 是 4 个模型的组合包，其 estimates 累加器、 shifts 累加器和内层输出会同时各持有一份完整副本，2 小时视频峰值约 34 GiB， 在 16 GB 机器上必然 OOM： RuntimeError: DefaultCPUAllocator: not enough memory: you tried to allocate 10442184992 bytes device=cuda 不能缓解，Demucs 只把逐段计算放到 GPU，完整音轨始终留在 CPU。 改为分块流式处理： - 先用 ffmpeg 把音频解码为临时 WAV，按窗口回读，避免重复解码 - 每个窗口多读 OVERLAP_SECONDS 的上下文，与上一窗口线性交叉淡化，接缝无损 - 人声与伴奏分别流式写入 float32 临时文件，再按全局峰值重缩放为 16-bit， 保持与 demucs save_audio 的 prevent_clip(rescale) 行为一致 - 新增 DEMUCS_CHUNK_SECONDS，默认 600 秒；内存约 0.28 GiB/分钟窗口长度， 默认约 2.8 GiB，与视频总长无关 实测：真实 htdemucs_ft 对 20 秒音频分 3 窗口处理，输出 882000 帧与输入 完全一致，临时文件清理干净。新增 11 个测试覆盖分块计划、交叉淡化、 ffmpeg 参数与端到端流式重建。 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Demucs 自带的加载器用 streams=0 明确读第 0 条音频流，而改成 ffmpeg 抽取后 只写了 -vn，ffmpeg 会按声道数与码率自动挑选"最佳"音频流。多音轨视频 （例如多语言配音的 YouTube 视频）可能因此换用另一条音轨。 加上 -map 0:a:0 与原行为对齐。 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
忽略无实际意义的语气词，纯语气词播放原始人声，补充音频合并回归测试。
Co-authored-by: traeagent <traeagent@users.noreply.github.com>
Co-authored-by: traeagent <traeagent@users.noreply.github.com>
Co-authored-by: traeagent <traeagent@users.noreply.github.com>
This was referenced Aug 26, 2026
Merged
Merged
liuzhao1225
added a commit
that referenced
this pull request
Aug 26, 2026
liuzhao1225
added a commit
that referenced
this pull request
Aug 26, 2026
Merged
liuzhao1225
added a commit
that referenced
this pull request
Aug 26, 2026
* 收录 Demucs 长音频分块与首音轨固定 - 按窗口流式分离并用重叠上下文交叉淡化，控制长视频内存占用 - 固定抽取第一条音频流，非法分块配置直接报错 - 清理成功及异常路径的临时文件和半成品，补齐回归与磁盘说明 - 背景：基于 PR #112 的 d86ad4c 与 6d5d10e 按当前 main 重做 Co-authored-by: lyk-bit <lyk-bit@users.noreply.github.com> * 修复 Demucs 跨块峰值与大文件边界 - 将单块读取、推理和写出收进独立作用域，下一块开始前释放完整张量 - 输入启用 FFmpeg RF64 自动切换，临时 stem 与最终输出固定使用 RF64 - 用 weakref 验证跨块释放，并补充真实磁盘公式与峰值保证 * 恢复 Demucs 首音轨声道语义 - 用 ffprobe 探测第一音轨声道数，mono 复制且多声道只取前两路 - 以 float32 和 RF64 自动策略解码，保持长音频精度与大文件能力 - 增加真实 5.1 独有频率与 mono 复制回归，更新磁盘空间公式 * 修复 Demucs 残件误用为缓存 - 每次实际分离先清理旧输出并完整重算 - 两份 pending 写完后再发布，异常时清除所有残件 - 补充恢复回归及 SIGKILL 遗留策略文档 --------- Co-authored-by: lyk-bit <lyk-bit@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.