Skip to content

CHANGELOG.md

Latest commit

 

History

History
266 lines (165 loc) · 25.2 KB

CHANGELOG.md

File metadata and controls

266 lines (165 loc) · 25.2 KB
Raw

更新日志 / Changelog

v2.9.3 (2026-08-09)

🔧 修复 / Fixes

  • 本地引擎安装链加固：下载失败会清理残留 .part，SHA-256 必须是严格的 64 位十六进制值；解压前完整验证成员，拒绝路径穿越、绝对路径、链接与设备文件。 Hardened local-engine installation: Failed downloads remove partial files, SHA-256 values must be exact, and archives are fully validated before extraction to reject traversal, absolute paths, links, and device entries.

  • 本地引擎信任与卸载边界：持久化 manifest 不再能覆盖内置下载地址或文件布局；卸载会清除运行时、模型、缓存、下载和状态，且不允许删除安装根目录之外的路径。 Bounded local-engine trust and uninstall: Persisted manifests cannot override bundled download sources or layout, and uninstall removes all provider artifacts without deleting outside its install root.

  • 用户数据写入与删除安全：设置、历史和声音缓存改为同目录原子替换；历史删除拒绝音频目录外路径，深度清理同时覆盖应用生成的 WAV。 Safer user-data writes and deletion: Settings, history, and voice cache use atomic replacement; history deletion is contained to the audio directory, and deep cleanup includes generated WAV files.

  • 缓存、许可证与工程维护：Kokoro 缓存身份现在包含语速、音量和音高；补全 GPL-3.0-only/NOTICE 及发布包携带，固定直接依赖，新增 Windows CI 与 Dependabot，并更新仓库所有者链接。 Cache, licensing, and maintenance: Kokoro cache identity now includes rate, volume, and pitch; GPL/NOTICE packaging, direct dependency pins, Windows CI, Dependabot, and repository-owner links are corrected.

v2.9.2 (2026-03-17)

🔧 修复 / Fixes

  • 换声音后再次播放会重新生成：播放按钮的缓存判定现在会包含 引擎 + 声音/音色(sid) + 语速/音量/音高 + 文本，避免“文本没变但换声音仍播放旧音频”的问题。 Voice switch now takes effect on next play: Playback cache validation now includes engine + voice/sid + rate/volume/pitch + text, preventing stale audio when switching voices without changing text.

  • 离线点读恢复（通过重新合成）：离线 Kokoro 无时间戳时，播放界面仍提供可点击文本覆盖层；点击某字/词会停止当前播放并重新合成从该位置开始的文本后播放（不写入历史、不复制到剪贴板）。 Offline point-read restored (via re-synthesis): When offline Kokoro has no timestamps, the UI still provides a clickable text overlay. Clicking a character/word stops current playback and re-synthesizes the suffix from that position, without writing history or copying to clipboard.

  • 离线点读忙时断路器 + 长文本保护：点读生成中会拒绝后续点击；文本过长时自动禁用点读覆盖层并提示，避免 UI 卡顿。 Busy-state circuit breaker + long-text guard for offline point-read: Rejects additional clicks while generating, and disables the overlay for very long text to prevent UI lag.

v2.9.1 (2026-03-16)

🔧 修复 / Fixes

  • 离线音色列表改为官方 Kokoro v1.1 目录：离线模式下声音列表不再显示微软 voice，而是按“美式女声 / 英式女声 / 中文女声 / 中文男声”分组展示 Kokoro sid，并新增“离线语音音效听”外链入口。 Offline voice catalog now matches official Kokoro v1.1: Offline mode no longer shows Microsoft voices. It groups Kokoro sid voices by US/UK English female and Chinese female/male, and adds an external “offline voice demos” link.

  • 修复引擎切换后仍播放旧音频造成误判：切换在线/离线时会停止播放、清空当前音频状态并提示重新生成，避免“选了在线但像本地在读”的错觉。 Fix stale playback after engine switch: Switching engines now stops playback and clears current audio state to avoid confusing stale audio with the new engine.

  • 点读/跳句在离线模式给出明确提示：离线引擎 V1 不产出时间戳，点读与跳句会提示“不支持”，在线模式则增加 seek 兼容 fallback，降低跳转失败概率。 Point-read / sentence jump now behaves predictably: Offline V1 has no timestamps so point-read/jump shows a clear “not supported” message. Online mode adds a seek fallback to reduce jump failures.

  • Edge 失败错误更可诊断：在线生成失败不再只显示 debug_edge_tts_internal_error，而是透传真实异常信息，便于定位 voice/网络问题。 More diagnosable Edge failures: Online errors now propagate the real exception instead of a generic debug_edge_tts_internal_error.

v2.9.0 (2026-03-16)

✨ 新功能 / Features

  • 离线引擎（Kokoro 本地边车）：新增 local_kokoro Provider，可在设置页一键下载/校验/卸载并切换在线/离线；下载链路强制 SHA256 校验并内置健康检查；安装目录固定到 %APPDATA%/Anki-TTS-Edge/providers/kokoro/。 Offline engine (Local Kokoro sidecar): Added local_kokoro provider with one-click download/validate/uninstall and online/offline switch from Settings. Downloads are SHA256-verified with built-in healthcheck, installed under %APPDATA%/Anki-TTS-Edge/providers/kokoro/.

  • TTSProvider 抽象层：统一主生成入口，并在返回结构中预留 timestamps（离线 V1 返回空，未来可无痛补齐）。 TTSProvider abstraction: Unified the generation pipeline and reserved timestamps in the return structure (offline V1 returns empty; future upgrades won't require main-logic refactors).

🔧 修复 / Fixes

  • 离线运行时更稳的默认下载项：默认下载 *-win-x64-*-MT-Release.tar.bz2 运行时包（含可用的 sherpa-onnx-offline-tts.exe），并通过 manifest version 自动升级旧 manifest，避免卡死/缺依赖导致不可用。 More reliable offline runtime default: Default download now uses the *-win-x64-*-MT-Release.tar.bz2 runtime bundle (includes a working sherpa-onnx-offline-tts.exe) and auto-upgrades older manifests by version to avoid hangs/missing-dependency failures.

  • 模型文件名兼容：自动识别 model*.onnx（如 model.int8.onnx），避免解压成功但校验失败。 Model filename compatibility: Auto-detects model*.onnx (e.g. model.int8.onnx) to prevent "extracted but not found" validation failures.

v2.8.2 (2026-03-13)

🔧 修复 / Fixes

  • 监听语义正式拆分：设置项现在区分“复制后生成音频”“鼠标划选生成音频”“双语音模式”“划词双语音模式”，首页双生成按钮与划词卫星双点不再共享同一个旧字段。 Monitoring semantics split cleanly: Settings now distinguish between copy-triggered generation, mouse-selection generation, dual voice mode, and selection dual voice mode. The home view A/B buttons and the selection satellite A/B dots no longer depend on the same legacy flag.

  • 设置页分组与文案重排：按“播放 / 声音模式 / 划词模式 / 复制模式 / 窗口 / 存储 / 维护”重排设置页，并将“鼠标划选生成音频”更名为“划词单语音模式”，与“划词双语音模式”并排。 Settings page regrouped and renamed: Reorganized settings into Playback, Voice Mode, Selection Mode, Copy Mode, Window, Storage, and Maintenance, and renamed the selection single-toggle to “Selection Single Voice Mode” so it sits next to the dual-selection toggle.

  • 修复 Ctrl+C 复制后只填文本不生成的问题：剪贴板监听现在会走统一生成入口，自动使用最新声音生成音频，并按自动播放设置决定是否立即朗读。 Fix copied text only filling the input without generating audio: Clipboard monitoring now goes through the unified generation pipeline, uses the latest voice automatically, and respects autoplay for immediate playback.

  • 修复生成后 MP3 文件到剪贴板失效：文件剪贴板写入改为带重试的 CF_HDROP 方案，并在写入前抑制内部监听，避免刚生成的 MP3 再反向干扰文本监听。 Fix broken “put MP3 on clipboard after generation” flow: File clipboard writes now use a retrying CF_HDROP path and suppress the internal monitor before writing so generated MP3 copies no longer feed back into text monitoring.

  • 修复历史页按钮失效：历史项外层不再吞掉“播放 / 删除”按钮事件，清空全部改回标准 page.dialog 链路。 Fix inactive history controls: The outer history item container no longer swallows Play/Delete clicks, and Clear All now uses the standard page.dialog flow.

  • 降低划词提示与监听延迟：划词优先尝试直接读取当前控件选区，失败时才退回 Ctrl+C，同时降低卫星与监听轮询延迟。 Reduce selection prompt latency: Selection capture now first tries direct control selection access before falling back to Ctrl+C, and both satellite and monitor polling have been tightened for faster response.

  • 修复划词与复制链路串扰：划词捕获不再复用剪贴板生成回调，内部选区读取和外部 Ctrl+C 生成被彻底拆分，避免未点 GO 自动生成以及系统剪贴板被划词流程污染。 Fix cross-talk between selection and clipboard flows: Selection capture no longer reuses the clipboard-generation callback. Internal selection reads and external Ctrl+C generation are now split cleanly, preventing auto-generation without GO and avoiding clipboard pollution from the selection flow.

  • 修复 A / B 与声音列表 1 / 2 的语义错位：双语音状态从“latest / previous”迁移为稳定的 left / right 槽位，首页和划词卫星统一固定为 A -> 声音列表 1、B -> 声音列表 2。 Fix A/B mapping drift against Voice List 1 / 2: Dual-voice state now uses stable left / right slots instead of the rotating latest / previous model, so both the home screen and the selection satellite now consistently map A -> Voice List 1 and B -> Voice List 2.

  • 新增运行时自检脚本：添加 tools/flet_runtime_selfcheck.py，在打包前验证视图构造、设置联动、历史页回调和当前 Flet 运行时签名，减少“进程活着但界面已经报错”的漏检。 Add runtime self-check script: Added tools/flet_runtime_selfcheck.py to validate view construction, settings coupling, history callbacks, and the active Flet runtime signature before packaging, reducing false positives where the process stays alive but the UI is already broken.

  • 修复划词模式仍会扰乱系统剪贴板：内部划词回退复制不再回放整份原始剪贴板句柄，而是只恢复用户真正可见的安全格式，降低文本复制/粘贴被污染的概率。 Fix selection mode still disturbing the system clipboard: Internal selection fallback no longer replays the entire raw clipboard payload and instead restores only safe user-visible formats, reducing the chance of corrupting normal copy/paste behavior.

  • 修复 MP3 文件到剪贴板“能写不能粘”：文件剪贴板改为标准 CF_HDROP + Preferred DropEffect 结构，并在本地验证可读回。 Fix MP3 clipboard writes that could not actually paste: File clipboard writes now use the standard CF_HDROP + Preferred DropEffect structure and were locally validated by round-trip reads.

  • 优化重复生成耗时：相同文本/声音/参数的请求现在直接命中本地音频缓存，避免重复走 edge-tts 生成。 Improve repeated generation latency: Identical text/voice/parameter requests now hit a local audio cache instead of regenerating through edge-tts.

  • 修复非“声音”页时后台功能像失效：复制后生成、划词生成、历史页播放/删除不再依赖当前 HomeView 已挂载，隐藏页状态更新统一走安全刷新链路。 Fix background features appearing broken outside the Voices tab: Clipboard generation, selection generation, and history play/delete no longer assume the Home view is currently mounted; hidden view updates now go through a safe refresh path.

  • 修复反复划词导致卡死和高占用：划词流程新增忙时断路器，在卫星窗等待点击或划词生成进行中时直接忽略新的划词触发，避免后台捕获与生成堆积。 Fix repeated selection causing freezes and high CPU/memory use: The selection flow now has a busy-state circuit breaker that ignores new selection triggers while the satellite is awaiting input or generation is already running.

  • 修复历史页右上角“清空全部”无响应：历史清空确认框切换到当前 Flet 运行时兼容的 page.open(dialog) / page.close(dialog) 链路，避免旧式 page.dialog 打开方式在打包版中看起来无反应。 Fix the unresponsive “Clear All” action in the history header: The history confirmation dialog now uses the runtime-compatible page.open(dialog) / page.close(dialog) flow instead of the legacy page.dialog pattern that appeared inert in packaged builds.

v2.8.1 (2026-03-13)

🔧 修复 / Fixes

  • 修复打包版启动时缺失 flet_desktop 运行时：将 flet 与 flet-desktop 固定到同一版本，并在 PyInstaller 构建中显式收集 flet_desktop 模块与资源，解决 EXE 启动后立即报 ModuleNotFoundError: No module named 'flet_desktop' 的问题。 Fix missing flet_desktop runtime in packaged builds: Pinned flet and flet-desktop to the same version and explicitly collected flet_desktop in PyInstaller so the EXE no longer fails on startup with ModuleNotFoundError: No module named 'flet_desktop'.

  • 修复主题配色与当前 Flet 运行时不兼容导致的启动异常：主题构造改为兼容模式，遇到旧版 ColorScheme 不支持的字段时自动降级，解决 surface_variant 触发的启动错误。 Fix startup error caused by ColorScheme incompatibility: Theme creation now falls back gracefully when older Flet runtimes do not accept newer ColorScheme fields, fixing the startup failure triggered by surface_variant.

  • 修复视图类写入只读 page 属性导致的启动异常：HomeView、HistoryView、SettingsView 改用私有页面引用，兼容当前 Flet 控件基类的只读属性约束。 Fix startup error caused by writing to the read-only page property: HomeView, HistoryView, and SettingsView now store the host page in a private reference instead of assigning to Flet's read-only page property.

  • 修复首屏和页面交互的强延迟：语音列表改为“优先本地缓存、命中缓存时再后台刷新”，顶部三页切换改为单宿主视图切换，托盘与卫星轮询改为惰性/降频启动，长文本播放不再无上限构建逐词高亮控件。 Fix severe UI latency on startup and page interactions: Voice loading now prefers local cache and only refreshes in the background when cache exists, the top navigation uses a single hosted view instead of a three-page stack, tray/satellite work is lazily initialized or throttled, and long playback sessions no longer build unbounded per-word highlight controls.

v2.8 (2026-03-13)

🔧 修复 / Fixes

  • 修复安装后缺依赖直接崩溃：补齐 pynput 与 pywin32 依赖，并同步更新打包入口，避免按 README 安装后因监听或文件剪贴板模块缺失而无法启动。 Fix startup crashes caused by missing dependencies: Added pynput and pywin32 to the runtime requirements and updated the packaging entry so fresh installs no longer fail before launch.

  • 修复托盘/卫星恢复主窗口失效：移除残留旧版 Flet 窗口 API，统一切换到 page.window.* 新接口，恢复托盘显示/隐藏与卫星双击唤醒链路。 Fix tray/satellite window restore flow: Removed legacy Flet window calls and standardized on the current page.window.* API so tray restore and satellite double-click bring the main window back reliably.

  • 修复历史回放状态串音：播放历史记录时现在会同步刷新文本与时间戳，避免沿用上一条音频的高亮、句导航与播放状态。 Fix state leakage when replaying history: History playback now refreshes the active text and timestamps so highlighting, sentence navigation, and playback state always match the selected record.

  • 修复历史缓存只删记录不删文件：达到历史上限时会同步删除被淘汰记录对应的音频和 .timestamps.json，磁盘占用不再无限增长。 Fix cache trimming leaving orphaned files: When history exceeds the configured limit, the app now deletes the evicted audio file and matching .timestamps.json instead of only trimming the JSON list.

  • 修复音频文件秒级重名覆盖：输出文件名改为“微秒时间戳 + 短 UUID”，高频连续生成不会再静默覆盖前一个文件。 Fix silent audio overwrite on rapid generation: Output filenames now use a microsecond timestamp plus a short UUID, preventing collisions during back-to-back generations.

  • 修复主题与设置持久化不一致：统一使用 appearance_mode 存储主题，并兼容旧版 theme_dark 设置迁移；同时解除剪贴板监听与划词监听的错误绑定。 Fix inconsistent settings persistence: Theme storage now uses a single appearance_mode key with compatibility migration from theme_dark, and clipboard monitoring is no longer incorrectly coupled to selection monitoring.

  • 修复播放监控并发竞态：播放、高亮与暂停恢复监控改为单实例运行，避免多个后台协程同时修改 UI 和播放状态。 Fix playback monitor race conditions: Playback/highlight monitoring is now single-instance, preventing multiple background tasks from mutating the UI and playback state at the same time.

📚 文档 / Docs

  • 新增架构文档：补充 ARCHITECTURE.md，固化主流程、状态边界、持久化规则与本轮修复决策，降低后续维护成本。 New architecture document: Added ARCHITECTURE.md to document the runtime flow, state boundaries, persistence rules, and the decisions behind this stabilization pass.

v2.7 (2026-03-01)

✨ 新功能 / Features

  • 快捷打开数据目录：在设置界面新增“打开数据目录”按钮（存储了音频、历史记录、配置文件等），方便高级管理。 Quick access to data directory: Added a new "Open Data Directory" button in Settings to easily access audio files, history, and config files.

v2.6 (2026-03-01)

✨ 改进 / Improvements

  • 中性配色主题：浅色/深色模式均改为 Slate 蓝灰色系，无强烈色彩倾向，视觉中性专业，适合所有用户。 Neutral color theme: Both light/dark modes now use a Slate blue-grey palette with no strong color bias — clean, professional, and universally comfortable.

  • 设置即改即存：移除"保存设置"按钮，所有设置项修改后立即生效并自动保存，减少不必要的点击。 Instant settings save: Removed the "Save Settings" button. All setting changes now take effect and are saved immediately, eliminating unnecessary clicks.

  • README 核心特性精简：核心特性小节不再堆积更新日志条目，仅保留软件功能的简洁描述。 README features cleanup: The Core Features section now contains only clean feature descriptions, not changelog entries.

v2.5 (2026-03-01)

✨ 新功能 / Features

  • 用户数据统一管理：所有用户生成的文件（设置、历史、音频、日志、语音缓存）迁移至 %APPDATA%/Anki-TTS-Edge/，目录使用英文命名（audio/、logs/），全球通用。首次启动自动迁移旧数据。 Centralized user data: All user-generated files (settings, history, audio, logs, voice cache) moved to %APPDATA%/Anki-TTS-Edge/ with English-named directories (audio/, logs/). Automatic one-time migration from old paths on first launch.

  • 深色/浅色主题重新设计：浅色模式采用暖色调 Teal 主色 + Stone 中性色，深色模式采用高对比 Teal-400 + Stone-900 背景，视觉更舒适自然。UI 控件全面使用语义化颜色 token（primary、surface、primaryContainer），主题切换无色彩断裂。 Theme redesign: Light mode uses warm Teal-600 primary + Stone neutral surfaces; dark mode uses vibrant Teal-400 + Stone-900 backgrounds. All UI components now use semantic color tokens for seamless theme switching.

v2.4 (2026-03-01)

✨ 新功能 / Features

  • 即点即读精确定位：点击文段中任意单词/字符时，直接从该位置开始播放，不再跳转到句首，定位更精确。 Precise click-to-play: Clicking any word/character in the text now starts playback from that exact position instead of jumping to the sentence start, providing more precise control.

📦 构建 / Build

  • 构建方式切换为 onedir 模式：从 PyInstaller 单文件 (onefile) 切换为文件夹 (onedir) 模式，大幅提升 Windows 下的启动速度（消除杀毒软件扫描导致的延迟）。 Switch to onedir build: Changed from PyInstaller single-file (onefile) to folder (onedir) mode, significantly improving Windows startup speed by eliminating antivirus scan delays.

v2.3 (2026-03-01)

🔧 修复 / Fixes

  • 修复 TTS 生成失败：升级 edge-tts 至 7.2.7，修复微软 API 认证令牌过期导致的 403 错误，恢复所有语音的正常生成能力。 Fix TTS generation failure: Upgraded edge-tts to 7.2.7, resolving 403 errors caused by expired Microsoft API authentication tokens. All voices are now functional again.

  • 修复单词高亮跟随失效：适配 edge-tts 7.x API 变更（boundary 参数默认值从 WordBoundary 改为 SentenceBoundary），显式传入 boundary='WordBoundary' 以恢复单词级时间戳和实时高亮。 Fix word-level highlight sync broken: Adapted to edge-tts 7.x API change where the boundary parameter default changed from WordBoundary to SentenceBoundary. Now explicitly passes boundary='WordBoundary' to restore word-level timestamps and real-time highlighting.

  • 修复重复函数定义 (handle_generate_a)：删除 main.py 中残留的重复函数定义。 Fix duplicate function definition (handle_generate_a): Removed leftover duplicate definition in main.py.

  • 修复窗口事件双重绑定：移除旧 API 的事件绑定 (page.on_window_event)，仅保留新 API (page.window.on_event)，避免窗口关闭/最小化事件被触发两次。 Fix double window event binding: Removed legacy API binding (page.on_window_event), keeping only the new API (page.window.on_event) to prevent duplicate event triggers on close/minimize.

  • 修复 MonitorManager.stop() 缺失：添加 stop() 方法作为 stop_monitors() 的别名，修复应用重启时的 AttributeError。 Fix missing MonitorManager.stop(): Added stop() as an alias for stop_monitors(), fixing AttributeError on app restart.

  • 修复卫星轮询中无效的 locals() 检查：satellite_loop 中的 monitor_manager 是闭包变量，locals() 永远无法检测到它，已改用 hasattr() 检查。 Fix invalid locals() check in satellite loop: monitor_manager is a closure variable, so locals() could never detect it. Replaced with hasattr().

⚡ 优化 / Improvements

  • 窗口缩放防抖：拖拽调整窗口大小时，设置文件不再每帧写入磁盘，改为 300ms 防抖延迟写入，大幅降低 I/O 压力。 Window resize debounce: Settings are no longer written to disk on every resize frame. Added 300ms debounce to significantly reduce I/O during window dragging.

  • 日志改为追加模式：剪贴板监控日志从覆盖模式 (filemode='w') 改为 RotatingFileHandler 追加模式，保留历史诊断信息，文件上限 1MB 自动轮转。 Log rotation: Clipboard monitor logging changed from overwrite mode (filemode='w') to RotatingFileHandler with 1MB cap, preserving diagnostic history.

  • 消除裸异常捕获：全项目 14 处 except: pass 替换为 except Exception，避免吞掉关键错误信息。 Eliminate bare excepts: All 14 instances of except: pass replaced with except Exception across the project to prevent silently swallowing critical errors.

  • 删除死代码：移除 audio_gen.py 中不再被调用的 generate_audio 和 generate_audio_edge_tts_async 函数及未使用的 threading 导入，代码行数减少 40%。 Remove dead code: Removed unused generate_audio and generate_audio_edge_tts_async functions along with the unused threading import from audio_gen.py, reducing file size by 40%.

v2.2 (2026-02-26)

  • 历史记录深度清理：自动扫描音频目录，清除孤立的音频文件和配套数据文件。 Deep history cleanup: Automatically scans audio directory and removes orphaned audio and metadata files.

v2.1

  • 高亮跟随朗读，即点即读，智能导航。 Real-time word highlighting, click-to-play, smart sentence navigation.

v2.0

  • 全新 Flet (Flutter) 重构，现代化 UI。 Full Flet (Flutter) rewrite with modern UI.