SmartSub: baixar versions anteriors

v3.11.0 「更多设置」全新分层，字体选择更省心，翻译输出更可靠
feat
- 「高级选项」升级为「更多设置」，分「常用」「专家」两页：常用项集中在前，解码、断句精调等专家项收进「专家」页；新建任务向导也新增了入口
Turn "Advanced options" into "More settings" with Common and Expert pages: everyday options come first and decoder-level knobs move to Expert; the new-task wizard gets an entry too
- 场景按钮在改过细项后仍显示场景名，并以「已调整」小圆点提示，弹层中可查看调整了哪些项并一键恢复；每个场景新增「适合什么视频」的说明
The scenario button keeps its name and shows an "Adjusted" dot after tweaks; the popover lists what changed with a one-click restore, and each scenario now says what kind of video it suits
- 任务工具栏按「听 → 译 → 出」重新排序，「字幕格式」改称「输出」；「字幕效果」更名为「识别策略」，其中的「自定义」档改称「手动调节」
Reorder the task toolbar as listen → translate → output and rename "Subtitle format" to "Output"; "Subtitle effect" is now "Recognition strategy", and its "Custom" tier is now "Manual tuning"
- 开始任务时，若视频自带文本字幕轨，会先询问一次：使用内封字幕，还是改用语音识别；可勾选记住选择、不再提示，之后在「更多设置 › 常用」中调整
When you start a task and a video carries a text subtitle track, SmartSub now asks once whether to use it or run speech recognition instead; tick the box to remember your choice and change it later in More settings › Common
- 字体选择器按「最近使用 / 常用字体 / 全部字体」分组，并支持用中文名和拼音（全拼或首字母）搜索字体
The font picker groups fonts into Recent / Common / All fonts and supports searching by Chinese name and pinyin (full or initials)
- 合成预览会提示哪些字体缺字、已临时改用了哪款字体，并推荐已安装的、能完整显示整段字幕的字体，点击即可切换
Compose preview now tells you when a font lacks glyphs and which font is used instead, and recommends installed fonts that can show the whole subtitle, switchable with one click
fix
- 修复 3.9.0 起选了「仅译文」字幕里仍带原文的问题：出厂默认恢复为仅译文，新建任务不再覆盖已保存的选择（此前已保存为双语的用户需自行改一次）；向导的合成面板也会说明成片字幕内容跟随该设置
Fix subtitles still carrying the source text after choosing translation only since 3.9.0: the factory default is translation only again and new tasks no longer override your saved choice (users who already saved bilingual need to switch once); the wizard's compose panel also says the final video follows this setting
- 修复个别情况下翻译任务显示成功却只得到空字幕文件，以及译文文件名与源字幕相同时会清空源字幕的问题，现在会明确报错并保留源字幕
Fix translation tasks occasionally reporting success while producing an empty subtitle file, and a translated file with the same name as the source subtitle wiping the source; these now fail with a clear error and the source is kept
- 修复在任务页调整的参数（含「更多设置」里的项）不运行任务就不会成为下一个新任务默认值的问题，现在改完即记住
Fix settings changed on the task page (including More settings) not becoming the defaults of the next new task unless the task was run; they are now remembered as you change them
- 修复字幕含韩文等所选字体不支持的字符时，合成预览和导出直接失败的问题：现在会自动改用可用字体继续，实在没有字体的字符显示为方块并给出提示
Fix compose preview and export failing outright when subtitles contain characters the selected font lacks (such as Korean): a fallback font is now used automatically, and characters no installed font covers show as boxes with a warning
- 降低 Qwen3-ASR 对长连续语音静默失败（输出空文本或「language」）的概率：默认将单段语音限制在约 15 秒内（最多 20 秒），仍超出解码预算时会在日志中给出警告 (#549)
Reduce the chance of Qwen3-ASR silently failing on long continuous speech (empty text or "language"): speech segments are capped at about 15 s by default (20 s at most), and a log warning is written when a segment still exceeds the decoding budget (#549)
- 修复向导和「在线下载后自动转写」的仅生成字幕任务，把字幕保存成 视频名_temp.srt 的问题，现在与任务页一致，保存为 视频名.srt
Fix subtitle-only tasks from the wizard and from download-then-transcribe saving subtitles as <video>_temp.srt; they now match the task page and save <video>.srt
- 修复在「手动调节」里改过的 VAD 等参数，改选内置识别策略后仍然生效、与所选档位不一致的问题
Fix parameters tuned under Manual tuning (such as VAD) still applying after switching back to a built-in recognition strategy
- 修复全新安装后场景按钮一开始就显示「专家自定义」的问题，现在默认显示「通用均衡」
Fix the scenario button showing "Expert Custom" on a fresh install; it now starts at "Balanced"

v3.10.1 崩溃诊断与稳定性加固，免费翻译更快更稳
feat
- 免费翻译提速：新增有道免费翻译源，并作为「自动免费翻译」的默认首选；谷歌免费翻译新增整批通道，被限流时自动退回逐条翻译（均为非官方接口，不保证长期可用）
Faster free translation: add Youdao as a free source and the default first choice of Auto Free; Google gains a batch lane that falls back to line-by-line when rate limited (both are unofficial endpoints with no availability guarantee)
- 翻译任务支持按任务选择术语表，并可选「生成摘要」：翻译前先通读全文生成摘要并注入 AI 翻译，提升人名与上下文的一致性 (感谢 @MaurUppi #436)
Choose glossaries per task, and optionally generate a summary before translating to keep names and context consistent in AI translation (thanks @MaurUppi #436)
- 新增崩溃诊断：「关于」页可导出诊断包，异常退出后也会主动提示（仅本地生成，不会自动上传）；导致闪退的加速后端会被自动停用，可在「GPU 加速」设置中查看并重试
Add crash diagnostics: export a diagnostics bundle from About, also offered after an abnormal exit (created locally, never uploaded automatically); backends that cause crashes are disabled automatically and can be retried in GPU acceleration settings
- 新增 CPU / 显卡兼容检测：CPU 不支持内置引擎（如缺少 AVX2）时给出提示并指引下载「旧 CPU 兼容包」；算力不足的 CUDA 加速包自动跳过，避免闪退
Add CPU / GPU compatibility checks: warn when the CPU can't run the built-in engine (e.g. no AVX2) with a link to the legacy CPU pack, and skip CUDA packs the GPU is too old for to avoid crashes
- 校对数据改存到应用数据目录，不再在视频旁生成 .smartsub-proofread 文件夹，并随任务一起清理（已有数据照常可用）
Proofread data now lives in the app data folder instead of a .smartsub-proofread folder beside the video, and is cleaned up with its task (existing data keeps working)
perf
- 内置引擎在 Metal / CoreML / CUDA 上启用 Flash Attention，转录更快
Enable Flash Attention for the built-in engine on Metal, CoreML and CUDA for faster transcription
- faster-whisper 语音复核在超长录音上明显提速，进度不再长时间停滞
Speed up faster-whisper speech review on very long recordings, with progress no longer stalling
- 任务列表界面优化：按状态着色描边，运行中各列保持对齐，拖入文件不再抖动，移除单个文件可撤销
Refine the task list: state-tinted outlines, stable alignment while running, no jitter when dragging files in, and undo when removing a single file
- 安装包体积缩小约 14%–18%：移除未被使用的依赖、重复拷贝与调试用的 source map，功能不受影响
Smaller installers (about 14–18% smaller): drop unused dependencies, duplicated copies and debug source maps, with no change in functionality
fix
- 修复 Windows 用户名或路径含中文等非 ASCII 字符时，内置引擎转录得到空字幕、语音检测/降噪/声音克隆失败的问题
Fix empty transcripts and failing voice detection, denoising and voice cloning on Windows when the user name or path contains non-ASCII characters
- 修复偶发的「write EPIPE」报错弹窗，以及 Windows 合成预览中鼠标移出后视频区域变黑的问题
Fix the occasional "write EPIPE" error dialog, and the video area turning black in the Windows compose preview after the pointer leaves it
- 修复烧录硬字幕时 ffmpeg 报「Conversion failed」：WebM 来源默认输出 MP4，兼容内嵌封面与奇数分辨率视频，失败时显示真实原因 (#521)
Fix "Conversion failed" when burning hard subtitles: WebM sources default to MP4, embedded cover art and odd-sized videos are handled, and the real failure reason is shown (#521)
- 提升 AI 断句通过率：容忍复述时的细微偏差、保留最佳结果、请求不再长时间卡住，失败时给出准确原因 (#507)
Improve AI segmentation success: tolerate small copy drift, keep the best answer, no more long-stalled requests, and report accurate failure reasons (#507)
- 修复校对台两个问题：含零时长字幕的校对数据无法打开（已有数据自动修复）；播放器字幕预览缺失或编辑后不更新 (#511 #505 #520)
Fix two proofread issues: data with zero-length cues failed to open (existing data is repaired automatically), and the player subtitle preview was missing or not following edits (#511 #505 #520)
- 修复替换同路径视频后仍沿用旧音频转录的问题 (#510)
Fix a replaced video at the same path being transcribed from stale cached audio (#510)
- 修复双语字幕配音时两种语言都被朗读的问题，现只朗读译文，导出字幕仍保留双语 (#537)
Fix dubbing reading both languages of a bilingual subtitle: only the translation is spoken now, and the exported subtitle stays bilingual (#537)
- 修复 faster-whisper 模型路径指向共享的 HuggingFace 缓存时，目录被擅自移动导致其他软件无法加载的问题；删除模型时会列出将删除的目录 (#519)
Fix faster-whisper moving folders in a shared HuggingFace cache (breaking other tools); deleting a model now lists the folders to be removed (#519)
- 修复新建任务不再沿用上次所用配置的问题
Fix new tasks no longer starting from the last used settings

v3.9.0 🤖 AI 创作助手登场，MCP 自动化与音视频工具箱
feat
- 新增内置 AI 创作助手（快捷键 ⌘J / Ctrl+J），支持上下文感知、对话式字幕润色、截图排障与文件对话 (#499)
Add in-app AI Assistant (⌘J / Ctrl+J) with context awareness, conversational subtitle polish, screenshot diagnostics, and file chat (#499)
- 支持 MCP 协议与 CLI 自动化，提供 111 个工具与统一命令行，支持一键连接 Cursor、Codex 与 Claude Code (#499)
Support MCP and CLI automation with 111 tools and unified command line, integrating with Cursor, Codex, and Claude Code (#499)
- 新增音视频工具箱，内置音频提取、字幕合并/转换/对齐、视频压缩、裁剪与转 GIF 等实用小工具 (#487)
Add Audio/Video Toolbox for audio extraction, subtitle merge/convert/sync, video compression, trimming, and GIF creation (#487)
- 新增字幕质量巡检，智能检测语速过快 (CPS)、单行超长与疑似漏听，支持快速试听与补全 (#496 #497)
Add subtitle quality review to detect rapid speech rate (CPS), overly long lines, and missed speech with quick audition and repair (#496 #497)
- 支持多视频独立匹配各自的参考文稿，全局文稿移至高级设置 (#488 #498)
Support independent manuscript matching for multi-video imports, and move global manuscript into advanced settings (#488 #498)
- faster-whisper 支持从本地压缩包或目录一键导入运行时，无需重新下载 (#495)
Support importing faster-whisper runtime directly from local archives or folders (#495)
- 任务列表新增微阶段进度与耗时展示，并将工具箱处理记录纳入历史任务 (#498)
Display micro-stage progress and elapsed time for tasks, and include toolbox operations in task history (#498)
perf
- 重构校对台布局并合并工具栏，优化宽窄屏自适应与长字幕列表流畅度 (#494 #497)
Refactor proofread layout and merge toolbars, improving responsiveness and list performance (#494 #497)
- 优化语速 (CPS) 计算逻辑，忽略空白字符并加入动态容差，减少误报 (#499)
Optimize CPS calculation by ignoring whitespace and adding dynamic tolerance to reduce false alarms (#499)
fix
- 修复自定义参数中嵌套对象显示为 [object Object] 的问题，支持编辑 JSON 对象并保留 thinking 配置 (感谢 @nightt5879 #477)
Fix nested parameter objects showing as [object Object], supporting JSON object editing and thinking configs (thanks @nightt5879 #477)
- 修复翻译服务测试在接口返回错误时被误判为连通成功的问题 (感谢 @lllleolin-max #481)
Fix provider test false positives when endpoints return structured errors (thanks @lllleolin-max #481)
- 修复 OpenAI 兼容翻译请求未走应用代理的问题，并在连接失败时停止无效探测 (#479 #483)
Fix OpenAI-compatible translation requests not using app proxy, and stop probing on failure (#479 #483)
- 翻译部分行失败时明确提示失败行数，并允许进入校对台查看与补译 (#489)
Show failed line count on partial translation errors and allow proofreading untranslated lines (#489)
- 优化 AI 字幕精修与翻译服务的配置联动，解决服务冲突与提示词保存问题 (#484 #485 #497)
Improve config coordination between AI refine and translation, resolving service conflicts (#484 #485 #497)
- 过滤 Qwen-ASR 输出中的特殊控制字符，防止字幕解析与渲染异常 (#494)
Sanitize control characters in Qwen-ASR output to prevent subtitle parsing and display errors (#494)
- 修复 Windows 下合成视频导出时偶发的 fsync EPERM 写入报错 (#494)
Fix intermittent fsync EPERM errors when exporting composed videos on Windows (#494)
贡献者 / Contributors
感谢 @nightt5879 贡献的嵌套对象参数与 Thinking 配置支持 (#477)
Thanks to @nightt5879 for contributing nested object parameters and thinking configuration support (#477)
感谢 @lllleolin-max 贡献的服务商测试失败状态准确上报修复 (#481)
Thanks to @lllleolin-max for fixing accurate failure reporting in provider connectivity tests (#481)

v3.8.0 多格式字幕导出、备用服务容灾与音频引擎扩容
💐 今天是教师节，向所有传道授业、启迪心灵的老师与教育工作者致以最真挚的敬意，祝各位老师节日快乐！
Happy Teachers' Day to all teachers and educators who inspire, guide, and share knowledge—thank you for your dedication!
feat
- 支持一次导出多种字幕格式，可在 SRT、VTT、ASS、LRC、TXT 中自由多选，无需重复识别或翻译 (感谢 @nightt5879 #471)
Support exporting multiple subtitle formats simultaneously (SRT, VTT, ASS, LRC, TXT) in a single run without redundant transcription or translation (thanks @nightt5879 #471)
- 支持配置同类型备用翻译服务，在当前服务失败或额度用尽时自动顺延切换，大幅提升批量翻译稳定性 (#428 #466)
Support backup translation providers with automatic failover on error or quota exhaustion, improving batch translation reliability (#428 #466)
- 新增小米澎湃 MiMo 语音识别 (ASR) 与语音合成 (TTS) 服务商支持 (#472)
Add Xiaomi MiMo speech recognition (ASR) and speech synthesis (TTS) providers (#472)
- 新增 Parakeet v2 英文及日文 CTC 转写模型，提供更多轻量本地离线识别选择 (#416 #470)
Add Parakeet v2 English and Japanese CTC transcription models for more lightweight local offline ASR options (#416 #470)
- 引入疑似漏听检测与预警机制，结合音频能量与 VAD 标记缺口，便于快速复核与试听排查 (#433 #469)
Introduce suspected missed speech detection and warnings, combining audio energy and VAD gap analysis for fast review and audio audition (#433 #469)
- 支持在任务设置中关闭内封字幕直提，当原视频自带字幕不准确时可强制重新语音识别 (#419 #457)
Add an option to disable embedded subtitle extraction, allowing forced speech recognition when embedded subtitles are inaccurate (#419 #457)
- 本地配音支持统一解析字幕与目标语言，智能应用对应的数字、日期与电话朗读规则 (#463)
Dubbing automatically resolves subtitle and target language to apply accurate local number, date, and phone reading rules (#463)
perf
- 调优 AI 翻译默认批量大小（20 条）与并发数，AI 语义断句复用服务商并发，显著提升处理速度 (#475)
Optimize AI translation default batch size (20 lines) and concurrency, with AI segmentation reusing provider concurrency for faster throughput (#475)
fix
- 修复 Kokoro 英文音色误加载中文规则导致数字朗读异常的问题 (感谢 @nightt5879 #458)
Fix Kokoro English voices incorrectly loading Chinese rules which caused abnormal number reading (thanks @nightt5879 #458)
- 改进 GPT-5.6 / Gemini 等模型的思考控制参数映射，并在端点拒绝自动参数时自动去参降级重试 (#438 #453 #455)
Improve thinking control parameter mapping for models like GPT-5.6 and Gemini, automatically retrying without thinking parameters if rejected by endpoints (#438 #453 #455)
- 纯音频输入时在任务向导和流水线中明确拒绝视频压制合成，避免 FFmpeg 因无视频流报错 (#465 #467)
Reject video composition for audio-only inputs in the task wizard and pipeline, avoiding FFmpeg stream-mapping errors (#465 #467)
- 更新 Bing 与 DeepLX 免费翻译服务协议及速率限制处理，保障免费翻译可用性 (#473 #474)
Update Bing and DeepLX free translation protocols and rate-limiting handling to restore free translation reliability (#473 #474)
- 修复使用 ElevenLabs Scribe v2 转写阿拉伯语时单词间空格丢失拼接的问题 (#449 #456)
Fix missing word spacing when generating Arabic SRT subtitles with ElevenLabs Scribe v2 (#449 #456)
- 净化 Sherpa Qwen-ASR 输出中的控制字符，修复特定环境下转写结果 JSON 解析崩溃的问题 (#427 #429)
Sanitize control characters in Sherpa Qwen-ASR output to fix JSON parsing crashes under specific environments (#427 #429)
- 将通义千问与 Qwen-MT 配置提示中的 API Key 获取链接更新为阿里云百炼控制台 (#460 #461)
Update DashScope and Qwen-MT API Key console links in provider tips to Alibaba Cloud Bailian console (#460 #461)
贡献者 / Contributors
感谢 @nightt5879 对本版本的贡献 (#458 #471)
Thanks to @nightt5879 for contributing to this release (#458 #471)
感谢 @FaintFlower 提交的文档修复 (#447)
Thanks to @FaintFlower for the documentation fix (#447)

v3.7.0 多角色配音与说话人识别，ASR 引擎再扩容
feat
- 新增 Qwen3-ASR 1.7B 转写引擎，语音识别精度与稳定性再升级 (#411)
New Qwen3-ASR 1.7B transcription engine for higher-accuracy, more reliable speech recognition (#411)
- 新增 Parakeet TDT 0.6B v3 转写引擎，本地转写又多一种轻量选择 (#413)
New Parakeet TDT 0.6B v3 transcription engine as another lightweight local option (#413)
- 支持手动选择 NVIDIA CUDA 显卡进行 GPU 加速，多卡环境可按需指定 (#412)
Choose a specific NVIDIA CUDA GPU for acceleration in multi-GPU setups (#412)
- 支持字幕说话人区分：自动识别不同发言者并标注角色 (#414)
Speaker diarization: automatically detect and label different speakers in subtitles (#414)
- 精修页支持角色命名、纠错与批量管理，多人对话字幕编辑更高效 (#422)
Proofread page: name, correct, and batch-manage speaker roles for multi-speaker dialogue (#422)
- 配音支持按角色批量分配音色，单句也可单独覆盖 (#423)
Dubbing: batch-assign voices by speaker role with per-line overrides (#423)
- 支持参考文稿高置信匹配：导入原稿可自动对齐并校正转写字幕 (#415)
Reference manuscript matching: import a script to align and correct transcription with high confidence (#415)
- 优化窗口管理与任务页布局，多窗口场景下界面更顺手
Improved window management and task page layout for smoother multi-window use
贡献者 / Contributors
感谢 @nightt5879 对本版本的主要贡献 (#411 #412 #413 #414 #415 #422 #423)
Thanks to @nightt5879 for the major contributions in this release (#411 #412 #413 #414 #415 #422 #423)

v3.6.0 ✨ AI 字幕精修登场，语言与转写更可控
feat
- 全新「断句与精修」：用 LLM 做语义断句与批量校正，工具栏统一入口，字幕可读性一步提升
New "Segment & Refine": LLM-powered semantic segmentation and batch correction from one toolbar entry for cleaner, more readable subtitles
- 支持自定义任务语言：在设置中添加名称与语言代码，源/目标语言下拉可选用 (#149) (感谢 @nightt5879 #406)
Custom task languages: add name + language code in Settings and use them in source/target language pickers (#149) (thanks @nightt5879 #406)
- 新增独立 Qwen-MT 机器翻译服务商，走官方翻译参数与术语，翻译更贴合通义机器翻译能力 (#376) (感谢 @nightt5879 #400)
New standalone Qwen-MT machine translation provider using official translation options and terms (#376) (thanks @nightt5879 #400)
- faster-whisper 高级设置新增多项解码参数（beam size、temperature 等），转写可按场景精细调优 (#294) (感谢 @nightt5879 #405)
New faster-whisper advanced decoding controls (beam size, temperature, and more) for finer transcription tuning (#294) (thanks @nightt5879 #405)
- 声音克隆参考音频转写增强：SenseVoice 不可用时自动回退到 faster-whisper / 内置模型 / 云端 ASR，成功率更高
Voice-clone reference transcription now falls back through faster-whisper, built-in models, and cloud ASR when SenseVoice is unavailable
fix
- 修复 AI 翻译偶发「原文原样回传」：自动检测未译条目并定点补翻，减少半成品译文 (#283) (感谢 @nightt5879 #404)
Fixed AI translations that sometimes echoed the source text: untranslated lines are detected and re-translated in place (#283) (thanks @nightt5879 #404)
- 修复统一存储目录未覆盖 Python runtime 的问题：切换存储根后 faster-whisper 运行环境也会跟随 (#401) (感谢 @nightt5879 #403)
Fixed unified storage root not covering the Python runtime: faster-whisper runtime now follows the storage root (#401) (thanks @nightt5879 #403)
- 修复 Deepgram 日语转写标点错位导致的异常断句与行首标点 (#391) (感谢 @nightt5879 #397)
Fixed Deepgram Japanese punctuation misplacement that caused bad sentence breaks and leading punctuation (#391) (thanks @nightt5879 #397)
- 修复 Windows 下视频下载完成后路径解析失败的问题，下载结果更稳定
Fixed Windows video-download path parsing failures after yt-dlp finishes, making download results more reliable

v3.4.0 配音与声音克隆登场
feat
- 全新「TTS 配音」功能：字幕一键生成配音，支持行级试听、调整与重新生成
New "TTS Dubbing" feature: turn subtitles into voiceovers with per-line preview, tweak and regenerate
- 配音服务商全家桶：Edge TTS、OpenAI 兼容、Azure、ElevenLabs、火山豆包，以及免联网的本地合成引擎
Full lineup of dubbing providers: Edge TTS, OpenAI-compatible, Azure, ElevenLabs, Volcano Doubao, plus an offline local engine
- 声音克隆：用一段参考音频或直接在向导内录音，克隆出自己的音色；支持本地 ZipVoice、火山复刻、ElevenLabs 即时克隆三种方式
Voice cloning: clone your own voice from a reference audio or in-wizard recording, via local ZipVoice, Volcano replication, or ElevenLabs instant cloning
- 音色管理更省心：内置降噪与质量检测、音色导入导出、云端音色误删可从平台取回
Easier voice management: built-in denoising and quality checks, voice import/export, and cloud voices recoverable after local deletion
- 全新界面：启动台仪表盘、垂直导航、剪辑软件风格的合并编辑器，配色与图标焕新
Brand-new UI: launchpad dashboard, vertical navigation, NLE-style merge editor, refreshed colors and app icon
- 任务向导支持保存「配方」：常用任务配置一键复用、可重命名管理
Task wizard now supports "recipes": save frequently used task configurations for one-click reuse
- 字幕合并（压制）支持硬件加速编码，并可保存个人字幕样式预设
Subtitle burn-in now supports hardware-accelerated encoding and personal style presets
- macOS Apple Silicon 新增加速方式设置（CoreML / Metal）
New acceleration options for Apple Silicon on macOS (CoreML / Metal)
- 日志中心升级：支持按日期与条件筛选查询，自动清理过期日志
Upgraded log center: filter logs by date and criteria, with automatic cleanup
perf
- 本地配音合成多进程并行，批量任务近线性提速
Local dubbing synthesis runs in parallel processes, near-linear speedup for batch tasks
- 本地语音引擎隔离运行，异常崩溃不再影响主程序
Local speech engines run in isolation — crashes no longer take down the app
fix
- 修复长静音场景下字幕句子错位的问题 (感谢 @nightt5879 #379)
Fixed subtitle misalignment after long silences (thanks @nightt5879 #379)
- 修复界面高度溢出与深色模式下文字对比度不足的问题
Fixed layout overflow and low text contrast in dark mode

feat
- 新增「云端听写」引擎：接入六家在线 ASR 服务商，免 GPU 即可转写
Added a "Cloud ASR" engine with six online providers — transcribe without a GPU
- 成句与断句增强：硬切词边界回退、三态断句设置（不拆词 / 可不限长度）
Smarter segmentation with word-boundary fallback and three-state line-break settings
- 任务高级设置支持可配置字幕重分割 (感谢 @nightt5879 #362)
Configurable subtitle resplitting in task advanced settings (thanks @nightt5879 #362)
- 校对行内新增单条删除按钮，支持撤销恢复 (#365)
Per-line delete in proofreading with undo support (#365)
- 批量翻译并发 / 批次 / 间隔三参数并排直显
Batch translation concurrency, batch size, and interval shown side-by-side
- 结构化输出测试时自动探测可用模式并写回配置
Structured output test auto-detects available modes and saves them to config
perf
- 任务并发调度改为阶段流水线：本地引擎仅转写互斥，云端按服务商全局限流
Stage-pipeline scheduling: local engines serialize transcription only, cloud providers get global rate limiting
- faster-whisper 变体切换免重复下载，秒级切回
faster-whisper variant switching without re-downloading, instant switch-back
fix
- ffmpeg 音频提取失败时显示具体错误原因 (#370)
ffmpeg audio extraction failures now show meaningful error messages (#370)

feat
- 内置 whisper.cpp 引擎全新细粒度时间轴：基于 Silero VAD 智能分段并设定最小显示时长，字幕断句与时间对齐更自然
Brand-new fine-grained timeline for the built-in whisper.cpp engine: Silero VAD based segmentation with a minimum display duration, for more natural subtitle timing
- 字幕效果档位：按使用意图自动派生复杂识别参数，几乎无需手动调参
Subtitle outcome presets: complex recognition parameters are derived from your intent, with little manual tuning
- 新增免费翻译源，支持多源自动回退与限速，翻译更稳定、更省成本
Added free translation sources with multi-source fallback and rate limiting, for more reliable and lower-cost translation
- 翻译支持批量并发配置，大批量字幕翻译速度更快 (感谢 @nightt5879 #344)
Configurable translation batch concurrency for faster large-batch translation (thanks @nightt5879 #344)
- 任务运行期间阻止系统休眠，长任务不再因息屏 / 休眠中断 (感谢 @nightt5879 #338)
Prevent system sleep while tasks are running, so long jobs are no longer interrupted (thanks @nightt5879 #338)
- 翻译服务商下拉按配置状态分组，已配置项优先展示，选择更高效 (#359)
Translation providers are grouped by configuration status with configured ones shown first, for quicker selection (#359)
perf
- 首屏空闲预取页面与语言包，页面切换更顺滑、几乎无加载等待 (#359)
Idle-time prefetch of pages and language packs, for noticeably smoother and near-instant navigation (#359)
fix
- 修复取消任务时流程阻塞、进行中的翻译请求未及时中止的问题 (感谢 @nightt5879 #353)
Fixed task cancellation hanging and in-flight translation requests not being aborted (thanks @nightt5879 #353)
- 增强 AI 翻译 JSON 响应解析的健壮性，减少因返回格式异常导致的失败 (感谢 @nightt5879 #345)
Hardened AI translation JSON response parsing to reduce failures caused by malformed responses (thanks @nightt5879 #345)
- 兼容误命名的字幕文件，自动探测其真实格式 (感谢 @nightt5879 #342)
Detect the real format of mis-named subtitle files (thanks @nightt5879 #342)
- 按语音结束时间收敛字幕显示，减少字幕滞留 (感谢 @nightt5879 #341)
Tighten subtitle display to the speech end time to reduce lingering captions (thanks @nightt5879 #341)
- 修复腾讯云翻译默认请求频率问题 (感谢 @nightt5879 #339)
Fixed the default request rate for Tencent Cloud translation (thanks @nightt5879 #339)
- TXT 校对入口在不支持时给出明确提示 (感谢 @nightt5879 #357)
Show a clear "unsupported" hint on the TXT proofreading entry (thanks @nightt5879 #357)

feat
- faster-whisper 支持 nvdia 显卡
faster-whisper supports nvdia graphics cards
fix
- 个别 2.x 用户升级之后无法执行仅生成原字幕任务
Some 2.x users cannot perform the task of only generating original subtitles after upgrading.

feat
- 字幕校对模块支持导入与导出多种字幕格式（SRT / VTT / ASS / LRC / TXT），并可在任务设置中选择输出格式
Subtitle proofreading now supports importing and exporting multiple subtitle formats (SRT / VTT / ASS / LRC / TXT), with a selectable output format in task settings
- 新增小牛翻译（NiuTrans）、腾讯云翻译、讯飞翻译三家翻译服务商
Added three new translation providers: NiuTrans, Tencent Cloud, and iFlytek
- 新增粤语（yue）源语言选项，提升粤语视频的语音识别准确度
Added Cantonese (yue) as a source language option for more accurate recognition of Cantonese videos
fix
- 修复部分语言码（如繁体中文 zh-hant）传入 Whisper 导致程序崩溃的问题，并在 CUDA 不可用时自动回退到 CPU
Fixed a crash when certain language codes (e.g., Traditional Chinese zh-hant) were passed to Whisper, with automatic CPU fallback when CUDA is unavailable
- 将 zh-cn / zh-tw / zh-hant / zh-hans 等中文变体统一映射为 zh，提升中文识别准确度
Unified all Chinese variants (zh-cn / zh-tw / zh-hant / zh-hans) to zh for better Chinese recognition accuracy
- 兼容新版 nvidia-smi 输出格式，修复部分 NVIDIA 显卡无法被检测、导致 GPU 加速不可用的问题
Compatible with the latest nvidia-smi output format, fixing GPU detection failures that disabled acceleration on some NVIDIA cards
- 修复字幕拆分弹窗内容过长时无法滚动、看不到确认按钮的问题
Fixed the subtitle split dialog so it scrolls when content is long and the confirm button stays visible
- 修复音频提取、字幕合并等进度在缺少百分比信息时不更新的问题（改用 timemark 推算进度）
Fixed progress not updating during audio extraction and subtitle merging when percentage info is unavailable (now derived from timemark)
- 校验 AI 翻译返回的条数，避免翻译结果与原文错位
Validate the count of AI translation results to prevent misalignment between translations and source text
- 为各翻译服务商的请求增加超时控制，避免单次请求卡死导致整体翻译进度停滞
Added request timeouts for translation providers to prevent a single stuck request from halting overall progress
- 修复字幕校对界面点击第一条字幕时，视频与列表跳转到错误位置的问题
Fixed the proofreading editor jumping the video and list to the wrong position when clicking the first subtitle

feat
- 导入视频支持选择文件夹，自动扫描文件夹及子目录中的所有媒体/字幕文件
Import supports selecting folders, automatically scanning all media/subtitle files in folders and subdirectories
- 新增硅基流动（SiliconFlow）翻译服务商
Added SiliconFlow translation provider
- 通义千问和 Deepseek 服务商配置支持自动获取可用模型列表
Tongyi Qianwen and Deepseek providers now support auto-fetching available model lists
- 任务列表显示文件数量，支持单条删除
Task list displays file count and supports deleting individual tasks
- 所有翻译服务商新增"请求间隔"设置，可自定义每次请求之间的等待时间以应对 API 限流
All translation providers now have a "Request Interval" setting for customizing wait time between requests to handle API rate limits
- GPU 加速包支持在轻量版和完整版之间切换，便于解决 CUDA 兼容性问题
GPU acceleration packages support switching between Lite and Full editions for resolving CUDA compatibility issues
- 新增配置导入导出功能，支持密码加密保护，方便跨设备迁移配置
Added configuration import/export with password encryption for easy cross-device migration
- 模型管理页面全新改版：按性能分类展示模型，根据系统内存推荐合适模型，优化交互体验
Redesigned model management page: categorized models by performance, system memory-based recommendations, improved UX
perf
- 翻译提示词优化：语言代码替换为完整语言名称（如 zh → 中文），提升翻译准确性
Translation prompt optimization: language codes replaced with full language names (e.g., zh → 中文) for better accuracy
- 升级时自动更新系统提示词，已自定义修改过的提示词不受影响
System prompts are automatically updated on upgrade; user-customized prompts are preserved
- 重构服务商配置代码，降低维护成本
Refactored provider configuration code for lower maintenance cost
- 模型下载重构为基于 Node.js 流的方式，支持断点续传、实时速度和剩余时间显示
Model download refactored to Node.js stream-based approach with resume support, real-time speed and ETA display
fix
- 修复 Mac 与 Windows 平台任务列表文件名显示不一致的问题
Fixed inconsistent filename display between Mac and Windows in the task list

v2.14 🌹🧨 情人节 × 春节特别版
🌹 情人节快乐，愿陪伴你的不止有工作，还有爱与温暖
🧧 新春快乐，万事如意！
从这个版本开始，显卡加速包迁移到软件应用内进行管理，需要在应用内进行下载，下载软件本体时无须再区分 cuda 版本号
🌹 Happy Valentine's Day, may there be more than just work accompanying you, there will also be love and warmth.
🧧 Happy New Year's Day, may all go well with you!
Starting from this version, the graphics card acceleration package will be migrated to be managed within the software application. It needs to be downloaded within the application, and when downloading the software 本体，there is no need to distinguish the CUDA version number.
feat
- 🎮 GPU 加速包应用内管理：无须手动安装 CUDA Toolkit，在设置中一键下载启用加速
🎮 In-app GPU acceleration pack management: No need to manually install CUDA Toolkit — download and enable from Settings
- 📁 自定义加速包路径：支持手动指定 addon.node 文件，灵活适配各种环境
📁 Custom acceleration pack path: Manually specify addon.node file for flexible environment support
fix
- 🐛 路径中含有特殊字符导致视频合并失败
🐛 The video merge failed due to special characters in the path.