Repository navigation
Releases
Releases · joeseesun/subcast
SubCast v0.2.9
中文
新增
- 英语学习模式(设置 › 学英语):
- 字幕里超出你水平的生词会标出来,点一下弹出词卡,有离线词典、音标,以及 AI 解释它在这句里的意思;
- 词卡上可以「加入生词本」或标为「认识」;
- 每个词保存原句和原声片段,复习时秒播;
- 按记忆曲线(FSRS)复习,有「回想」和「拼写」两种方式。拼写时有按键和对错音效,拼对了会撒花;
- 生词本有学习热力图、连续天数和记住率,每个词有详情卡片,点出处能跳回原节目对应的句子;
- 导出 Anki(.apkg,带原声)或 CSV,导出时先选保存位置;
- 本期生词、跟读、听写。
- 在 SubCast 里登录 YouTube:有些视频要求验证「不是机器人」,现在可以直接在 SubCast 里登录,不用再借浏览器的登录。这也解决了 Windows 上新版 Chrome 报「Failed to decrypt with DPAPI」的问题。
- 发现页:直接点 + 就能订阅;新增一个设置,可以隐藏侧栏里的乔木精读推荐。
改进
- 应用内更新更快:在我们的服务器和 GitHub 之间自动选更快的来源下载。新版本下好后,设置旁边出现升级按钮,点一下安装并重启。
- 字幕操作不再打断播放:选中、复制、翻译字幕不会跳转,也不会开始播放。单击一行仍然从这句开始播放。
- macOS 安装窗口:App 在左、Applications 在右,带背景和拖拽提示。
修复
- 云端转写的英文节目,英语学习模式不生效。
未验证
- Windows 和 Linux 版本由 CI 构建,没有在真机上完整测试。
- 在 SubCast 里登录 YouTube 的流程,以及 Windows 上的 DPAPI 修复,没有在 Windows 真机上验证。
升级
- 0.2.8 会在 App 里自动更新。
English
New
- English learning mode (Settings › Learn English):
- Words above your level are marked in subtitles. Click one for a card with an offline dictionary, IPA, and an AI note on what it means in this sentence.
- From the card, add the word to your word book or mark it as known.
- Each saved word keeps its sentence and a clip of the original audio, which replays instantly.
- Spaced review (FSRS) in two modes, recall and spelling. Spelling has key and right/wrong sounds, and confetti when you get a word right.
- The word book shows a study heatmap, your streak and recall rate. Each word has a detail card, and its source jumps back to that sentence in the episode.
- Export to Anki (.apkg with audio) or CSV; you choose where to save first.
- Words in this episode, shadowing and dictation.
- Sign in to YouTube inside SubCast: some videos ask you to confirm you're not a bot, and you can now sign in right in SubCast instead of borrowing a browser's sign-in. This also fixes "Failed to decrypt with DPAPI" with recent Chrome on Windows.
- Discover: subscribe in place with +, and a setting to hide the 乔木精读 picks in the sidebar.
Improved
- Faster in-app updates: the update downloads from whichever of our server and GitHub is faster. Once it's ready, an upgrade button beside Settings installs it and restarts.
- Subtitle actions no longer interrupt playback: selecting, copying or translating subtitle text neither jumps nor starts playback. A click on a line still plays from it.
- macOS installer window: SubCast on the left, Applications on the right, with a background and a drag hint.
Fixed
- English learning mode didn't work on English episodes transcribed in the cloud.
Not verified
- Windows and Linux builds come from CI and have not been fully tested on hardware.
- Signing in to YouTube inside SubCast, and the DPAPI fix, have not been tried on a Windows machine.
Upgrading
- 0.2.8 updates inside the app.
SubCast v0.2.8
中文
新增
- 音量控制:播放条上的喇叭按钮,鼠标移上去出现滑块,滚轮也能调;音频、视频和 YouTube 用同一个音量,下次打开还在。音量按人耳的感受变化,调一半听起来就是一半。快捷键:⌘↑ / ⌘↓ 或 = / −,M 静音。
- 从浏览器导入 YouTube 订阅时,SubCast 会先向 macOS 申请读取浏览器数据,弹窗里点「允许」即可。如果拒绝了,会说明怎么打开,并给出直达「文件和文件夹」的按钮,回来后自动重试。
改进
- 播放更快、更稳:
- 打开一集就开始缓冲音频。
- 禁止嵌入的 YouTube 视频会被记住,下次直接直连播放。
- YouTube 播放器加载不了时,12 秒后自动改走直连。
- 视频地址中途失效,会自动换新地址接着播。
- 转写长节目的内存大幅下降:48 分钟的节目从约 1.2 GB 降到约 200 MB。
- 播放时字幕列表、拖动分隔条、AI 问答流式输出都更顺,不再整页重绘。
- 节目内搜索不再卡界面,现在也能搜到节目简介里的内容。
- 刷新订阅时 6 个同时进行,单个慢的源最多等 30 秒,不再拖住全部。
- 云端识别取消或出错会立刻停止,不再把剩下的分段全部发出去。
修复
- 点播放后按钮一直转圈:播放出错、切换直连或改用音频时都会停下。
- 一边听节目一边打开另一个视频页,播放/暂停被那个视频抢走,离开后按钮失灵。
- 取消转写后马上重试,旧任务会删掉新任务的文件。
- 本机有其他程序占用 SubCast 的界面端口时,窗口会显示别的页面或一片空白。现在会自动换一个空闲端口。
未验证
- Windows 和 Linux 版本由 CI 构建,没有在真机上完整测试。
- 起播速度的提升是逐项改进,没有做前后对比测量。
升级
- 0.2.7 会在 App 里自动更新。
English
New
- Volume control: a speaker button on the playback bar with a slider on hover; the scroll wheel works too. Audio, video and YouTube share one level, kept across launches. The level follows how loudness is heard, so halfway sounds like half. Shortcuts: ⌘↑ / ⌘↓ or = / −, M to mute.
- Importing YouTube subscriptions from a browser: SubCast asks macOS for access to the browser's data first, so you only need to click Allow. If access is refused, it explains what to do, with a button straight to Files and Folders, and tries again when you come back.
Improved
- Faster, steadier playback:
- Audio starts buffering as soon as an episode opens.
- YouTube videos that refuse embedding are remembered and play directly next time.
- When the YouTube player can't load, playback switches to the direct stream after 12 seconds.
- A video address that expires mid-play is replaced automatically.
- Transcribing long episodes uses far less memory: a 48-minute episode went from about 1.2 GB to about 200 MB.
- Smoother subtitle list during playback, divider dragging and streamed AI answers, with no full-page redraws.
- Searching inside a show no longer freezes the window, and it now finds words in show notes.
- Refreshing subscriptions checks six feeds at a time, with at most 30 seconds per feed, so one slow feed no longer holds up the rest.
- Cloud recognition stops at once on cancel or error instead of sending the remaining chunks.
Fixed
- The play button kept spinning after a playback error, a switch to the direct stream, or a switch to audio.
- Opening another video's page while an episode played took over play/pause, and the controls stopped working after leaving that page.
- Retrying right after cancelling a transcription let the old run delete the new run's files.
- When another program held SubCast's interface port, the window showed someone else's pages or stayed blank. SubCast now moves to a free port.
Not verified
- Windows and Linux builds come from CI and have not been fully tested on hardware.
- The faster playback start comes from individual fixes and has not been measured before and after.
Upgrading
- 0.2.7 updates inside the app.
SubCast 0.2.7
中文
新增
- 别人分享过的字幕直接出:打开单集时先查本地、节目自带、共享字幕库,都没有才需要识别;共享库里有的,不用下载模型、不用点任何按钮。首页把有字幕的单集排在前面,并标「有字幕」。
- 模型装好后,按下播放就自动开始识别没有字幕的单集。
- 导读移到右侧标签,本集简介默认展开;导读也能共享:别人生成过的直接显示,不花你的额度。
- 收听统计重新设计:一句话总结你的收听、你把时间给了谁、收听生物钟、小发现、话题版图、耳朵的成长(不靠字幕的比例),以及可选的 AI 收听画像。
- 本机 AI(Codex、Kimi、Qwen、Gemini 等)可以读取自己支持的模型,勾选后出现在问答的模型选择里。
- 设置合并成 4 个区:字幕 / AI / 通用 / 关于。一个默认模型同时管问答、导读和翻译,分开指派放进高级设置;朗读设置挪到朗读按钮旁;⌘/ 查看快捷键。
修复
- 导读和问答偶尔出现「��」乱码。
- 导读跑完扣了额度却不显示;问答内容一会儿有一会儿没;切换标签不再丢失进行中的回答。
- 推理模型(如 DeepSeek V4.1 Flash)把篇幅用在思考上时,不再一片空白,会说明原因;兼容更多模型的返回格式。
- 已保存的 API Key 显示清楚;展开设置项时自动滚到展开的内容;字幕语言改用自己的菜单;术语表不再横向滚动。
- 双语字幕在视频上共用一个半透明底;默认「描边,无底」和思源黑体。
未验证
- Windows 和 Linux 版本由 CI 构建,没有在真机上完整测试;Windows 版还没有代码签名。
- 各家模型的返回格式只做了单元测试,没有逐一真机调用。
升级
- 0.2.5 会在 App 里自动更新。老用户的设置会自动迁移:选过谷歌翻译的仍用谷歌,关过自动翻译的改为只显示原文。
English
New
- Subtitles someone already shared appear right away: an episode checks its cache, the publisher's subtitles and the shared library before asking you to transcribe; no model download, no clicks. The home page lists episodes with subtitles first and marks them.
- Once a model is installed, pressing play starts transcribing episodes without subtitles.
- The guide moves to a tab on the right and show notes open by default; guides are shared too, so one someone made costs you nothing.
- Listening stats, redesigned: a one-line summary, where your time went, your listening clock, small discoveries, a topic map, subtitle-free share over time, and an optional AI portrait.
- Local agents (Codex, Kimi, Qwen, Gemini…) can list their models; ticked ones appear in the Q&A model picker.
- Settings in four sections: Subtitles / AI / General / About. One default model for Q&A, guides and translation (split in advanced); read-aloud settings next to the read-aloud button; ⌘/ shows shortcuts.
Fixed
- Occasional "��" in guides and answers.
- A finished guide that never showed; answers that vanished and came back; switching tabs no longer loses a reply in progress.
- Reasoning models that spend the length limit thinking now say so instead of a blank; more reply formats are understood.
- Saved API keys read clearly; expanding a setting scrolls it into view; own subtitle-language menu; no sideways scrolling in the glossary.
- Bilingual subtitles on video share one backdrop; outline and Source Han Sans are the defaults.
Not verified
- Windows and Linux builds come from CI and have not been fully tested on hardware; Windows is not code-signed yet.
- Reply formats of individual model providers are covered by unit tests, not live calls to each.
Upgrading
- 0.2.5 updates in the app. Settings migrate: Google translation stays Google; turning auto-translate off becomes original-only subtitles.
SubCast 0.2.5
中文
新增
- 收听统计:累计收听时长、听过的单集、连续收听天数、借助字幕听懂的时长,全年收听热力图、常听节目、来源和收听时段。只统计真正播放的时间,数据只存在本机。
- 设置 › 阅读与字幕:字幕字号和字体(思源宋体、霞鹜文楷、思源黑体)、默认显示双语 / 中文 / 原文、视频字幕底色、音频节目是否提前显示下一句、问答字号、快进快退秒数、播完自动接着播下一集。带实时预览。
- 第一次打开直接进首页,自动订阅几档精选节目,不再需要先挑节目。
修复
- 作者禁止嵌入的 YouTube 视频一直停在「正在获取视频地址…」:现在会使用系统代理(Clash 等),并找到本机的 JavaScript 运行环境;遇到「确认不是机器人」时可以在播放器里一键借用浏览器登录再试。
- 视频开始播放后,暂停按钮不能点。
- 点播放后有时不开始播放;等待时播放按钮会显示加载转圈。
- 视频字幕按画面大小缩放,全屏时比例不变。
- 音频字幕中英两行更紧凑,数字里的小数点不再被当成句号。
- 其他软件下载的 Qwen3-ASR 如果格式不能共用,设置里会说明原因。
未验证
- Windows 和 Linux 版本由 CI 构建,没有在真机上完整测试。
升级
- 0.2.4 会在 App 里自动更新。
English
New
- Listening stats: total listening time, episodes, streak, time with subtitles, a year heatmap, top shows, sources and listening hours. Only real playback counts; data stays on this computer.
- Settings › Reading & subtitles: subtitle size and font, default language, video subtitle look, next line on audio, Q&A text size, skip times and auto-play next, with a live preview.
- First launch opens the home page with a few curated shows subscribed, no picking screen.
Fixed
- YouTube videos that disallow embedding stuck on "getting the video address": the system proxy (Clash etc.) and a local JavaScript runtime are now used; YouTube's bot check can be passed with one click using your browser sign-in.
- Pause didn't work after a video started.
- Play sometimes didn't start; a spinner shows while waiting.
- Video subtitles scale with the picture, including full screen.
- Tighter bilingual audio subtitles; decimals no longer split sentences.
- Settings explains when another app's Qwen3-ASR can't be shared.
Not verified
- Windows and Linux builds come from CI and have not been fully tested on hardware.
Upgrading
- 0.2.4 updates inside the app.
SubCast 0.2.4
中文
新增
- 用你已有的 AI 订阅:自动找到本机的 Claude Code、Codex、Antigravity CLI(个人 Google 账号)、Qwen Code、Kimi Code、OpenCode,已登录的直接用来问答和翻译,不需要 API Key。它们保持运行,后面的回答更快;只能回答文字,不能读写你的文件。
- 模型设置重新设计:最上面写清楚现在用的是什么;本机 AI 和 API 模型分开,带图标和状态;没登录时告诉你在终端输入什么。
- 导入订阅:从浏览器一键导入 YouTube 订阅的频道;小宇宙和其他播客 App 导入 OPML 文件;也支持 Google Takeout、NewPipe、FreeTube。还可以导出 OPML。
- 视频全屏(F 键或双击,Esc 退出),双语字幕跟着放大。
- 快捷键可以在 设置 › 快捷键 里改,默认和 YouTube 一致(K 暂停、J/L 快退快进、M 静音、C 切字幕)。
- 语音识别新增 DeepSeek(和问答共用一个 Key);转写中途可以换一种识别方式。
- 其他软件下载过的识别模型(Screen Studio 等)直接拿来用,不用再下载。
改进
- 换页面后视频继续播放,缩到右下角小窗。
- 音频节目的字幕按句显示,下一句提前淡淡露出,句子切换时界面不跳动。
- YouTube 要求「确认不是机器人」时会自动换方式重试,也可以在 设置 › 通用 里借用浏览器登录。
未验证
- Windows 和 Linux 版本由 CI 构建,没有在真机上完整测试;本机 AI 和浏览器导入在 Windows 上未测试。
升级
- 0.2.3 会在 App 里自动更新。
English
New
- Use the AI subscriptions you already have: Claude Code, Codex, Antigravity CLI (personal Google accounts), Qwen Code, Kimi Code and OpenCode are found on this computer and, when signed in, answer questions and translate with no API key. They stay running so later answers come faster, and they can only reply with text, with no access to your files.
- Redesigned model settings: what is in use now at the top; local AI and API models side by side with icons and status; sign-in help when needed.
- Import subscriptions: YouTube channels from your browser in one step; Xiaoyuzhou and other podcast apps via OPML; Google Takeout, NewPipe and FreeTube too. Export OPML.
- Full-screen video (F or double-click, Esc to leave) with enlarged bilingual subtitles.
- Customisable shortcuts in Settings › Shortcuts, YouTube-style by default (K pause, J/L skip, M mute, C subtitles).
- DeepSeek speech recognition (shares the Q&A key); switch recognition engines mid-way.
- Speech models other apps already downloaded (Screen Studio and others) are used without downloading again.
Improved
- Video keeps playing in a small window when you leave its page.
- Audio subtitles show one sentence at a time with the next one faded in below, without the layout jumping.
- When YouTube asks to confirm you're not a bot, other ways are tried automatically; you can also allow a browser's sign-in in Settings › General.
Not verified
- Windows and Linux builds come from CI and have not been fully tested on hardware; local AI and browser import are untested on Windows.
Upgrading
- 0.2.3 updates inside the app.
SubCast 0.2.3
中文
改进
- 识别更准:本机识别(Qwen3-ASR)和豆包识别现在会参考节目标题和简介里的人名、公司和术语。实测一期中文节目,节目名和嘉宾名的错字从 7 处降到 0 处。之前本机 Qwen3 其实没收到任何节目信息。
- 翻译更准:翻译前先整理这一期的概要、说话人和术语表,每一句都照同一套译法翻,人名和术语前后一致。术语表可以在字幕栏里查看和修改。
- 翻译更稳更快:模型漏句或回答被截断时只补译缺的那几句,不再整批作废;先翻你正在听的位置;遇到服务限流会自动放慢、再恢复。
- 新增「精翻」:多一遍润色,读起来更像中文口语,花费约为两倍。在 设置 › 概览 › 字幕翻译 里切换。
- 共享译文:开启共享时,翻译完的一集会连同字幕一起匿名共享,下一个听这集的人打开就是双语。
未验证
- Windows 和 Linux 版本由 CI 构建,没有在真机上完整测试。
English
Improved
- More accurate recognition: on-device Qwen3-ASR and Doubao now use the names, companies and terms in the episode's title and show notes. On a Chinese episode, misspellings of the show and guest names went from 7 to 0. On-device Qwen3 previously received no episode information at all.
- More accurate translation: before translating, a short summary, the speakers and a glossary are prepared, and every line follows the same renderings. View and edit the glossary from the subtitle panel.
- Steadier, faster translation: when a model drops lines or its answer is cut off, only the missing lines are retried; the part you are listening to goes first; rate limits slow it down and it speeds back up.
- New "Best" quality: a second polishing pass for more natural Chinese, about twice the cost. Settings › Overview › Subtitle translation.
- Shared translations: with sharing on, a finished translation is shared anonymously with the subtitles, so the next listener gets bilingual subtitles at once.
Not verified
- Windows and Linux builds come from CI and have not been fully tested on hardware.
SubCast 0.2.2
中文
新增
- 在节目里搜索:节目页「全部单集」旁的搜索框(⌘F),能搜标题和中文标题、节目简介,以及已有字幕的原文和中文。点字幕里的匹配,从那一句开始播放。
- 切换字幕语言:YouTube 视频有多种字幕时,可以在字幕栏右上角切换。
- 外语视频如果有节目方上传的中文字幕,双语里的中文那一行直接用它,比机器翻译更准。
- AI 问答默认只列你最常用的 4 个问题,其余收在「更多」里。
- 界面小修复以后不用等新版本:App 会在后台取到修复,下次打开时生效;万一修复本身有问题,会自动退回上一个能用的界面。
修复
- 中文视频打开却显示英文字幕(节目方同时上传了英文字幕时)。之前存下的英文字幕,打开时会自动换回中文。
- 英文节目在「精读」「问答」标签下,视频上的字幕只有英文:现在不管打开哪个标签都会翻译。
- YouTube 视频一直停在「正在切换到直连播放…」:现在显示进度,最多等 3 分钟,随时可以「在 YouTube 打开」。播放组件从国内可访问的镜像下载,并校验文件。
- 不允许嵌入的视频在 Windows 上无法播放:现在可以播放,画质最高 1080p。
- Windows 打开时提示「找不到 MSVCP140.dll」:不再需要单独安装 Visual C++ 运行库,本机识别(Qwen3-ASR)也一样。
- Windows 上快捷键显示为 Ctrl,而不是 ⌘。
未验证
- Windows 和 Linux 版本由 CI 构建,没有在真机上完整测试。
- Windows 版本尚未签名,首次打开时 SmartScreen 会提示,选「更多信息 → 仍要运行」。
升级
- 0.2.0 会在 App 里自动更新。
English
New
- Search inside a show: the box next to All episodes (⌘F / Ctrl+F) searches titles and their Chinese, show notes, and subtitles of episodes that have them. Click a subtitle match to play from that line.
- Switch subtitle language when a YouTube video has captions in several languages.
- When a foreign-language video has the publisher's own Chinese subtitles, the bilingual view uses them instead of machine translation.
- AI Q&A lists your four most-used questions; the rest are under More.
- Small interface fixes no longer wait for a new version: the app picks them up in the background and uses them from the next launch, and goes back to the last working interface if a fix itself is broken.
Fixed
- Chinese videos opened with English subtitles when the publisher also uploaded English ones. Subtitles saved that way switch back to Chinese when opened.
- Subtitles on the video stayed English-only unless the Subtitles tab was open; translation now starts whichever tab is open.
- YouTube videos stuck on "switching to direct playback…": progress is shown, it gives up after 3 minutes, and Open on YouTube is always there. The playback component downloads from a mirror reachable in mainland China and is checksum-verified.
- Videos that don't allow embedding failed to play on Windows; they now play, up to 1080p.
- Windows: "MSVCP140.dll was not found" on launch. The Visual C++ Redistributable is no longer needed, including for on-device transcription (Qwen3-ASR).
- Shortcut labels show Ctrl instead of ⌘ on Windows and Linux.
Not verified
- Windows and Linux builds come from CI and have not been fully tested on hardware.
- Windows builds are not code-signed yet; on first launch choose "More info → Run anyway".
Upgrading
- 0.2.0 updates inside the app.
SubCast 0.2.0
中文
新增
- 首页重新设计:继续收听、接下来听、最新单集卡片;「为你推荐」按你在本机的订阅、收听、收藏和下载推荐新节目;宽屏左右两栏。
- 收藏单集,「继续收听」里可以一键按顺序连播;节目页也能「播放全部」。一集播完自动接下一集。
- AI 问答:回答可以带时间戳复制、去掉时间戳复制,或复制成时间戳跳到 YouTube 的 Markdown;可以重新回答、编辑后重问、新开对话并找回历史对话。
- AI 导读重写:要点更具体,标注说话人,覆盖整期,并推荐「先听这几段」。
- 金句海报:5 套主题和版式,5 款开源中文字体(思源宋体、霞鹜文楷、京华老宋体、思源黑体)。
- 精读频道显示完整节目简介,章节时间和链接可以点击。
- 离线下载与「存储」设置:查看占用、打开文件夹、清理。
- 自动更新:后台下载新版本,重启生效,可在「关于」里关闭。
- 四种界面语言:简体中文、繁體中文、English、日本語。
- 共享字幕:别人识别过的单集打开即有字幕;可在 设置 › 通用 关闭上传。
修复
- 侧栏菜单偶尔点了没反应。
- 取消订阅、删除下载点了没反应。
- 精读频道加载慢或报「error sending request」:现在并行加载、自动重试。
未验证
- Linux 版本由 CI 构建,没有在真机上完整测试。
- Windows 版本尚未签名,首次打开时 SmartScreen 会提示,选「更多信息 → 仍要运行」。
升级
- 这是第一个公开版本,从官网或本页下载安装。之后的版本会在 App 里自动更新。
- 协议改为 AGPL-3.0。
English
New
- Redesigned home: continue listening, up next, new episode cards, and For you recommendations based on what you follow, play, save and download on this computer. Two columns on wide windows.
- Save episodes and play them in order; Play all on show pages. The next episode starts when one ends.
- AI Q&A: copy answers with or without timestamps, or as Markdown whose timestamps link to YouTube; retry, edit and ask again, start a new chat and bring back earlier ones.
- Rewritten guide: more specific points, speakers named, the whole episode covered, and the parts worth hearing first.
- Quote posters: five themes with their own layouts and five open-source Chinese fonts.
- Full show notes for 精读 channels, with clickable chapter times and links.
- Offline downloads and a Storage page: sizes, open folder, clean up.
- Auto-update: new versions download in the background and apply on restart.
- Four interface languages: Simplified Chinese, Traditional Chinese, English, Japanese.
- Shared subtitles: episodes someone already transcribed open with subtitles ready. Uploading can be turned off in Settings › General.
Fixed
- Sidebar items sometimes ignored clicks.
- Unsubscribe and delete download did nothing.
- 精读 channels loading slowly or failing with "error sending request": requests now run in parallel and retry.
Not verified
- Linux builds come from CI and have not been fully tested on hardware.
- Windows builds are not code-signed yet; on first launch choose "More info → Run anyway".
Upgrading
- This is the first public release; download it from the website or this page. Later versions update inside the app.
- The license is now AGPL-3.0.