浏览器本地学习工具Browser-local learning tool· 2026
Sound NotebookSound Notebook
声音手帖 · 日语假名学习Japanese kana practice
认识、听辨、主动回答，再按记忆情况安排下一次复习。Recognize, listen, recall, then schedule the next review from an explicit memory rating.
案例摘要Case summary
目标用户Audience
希望用碎片时间练习日语假名的初学者；工具界面为中文。Beginners practicing Japanese kana in short sessions. The tool interface is in Chinese.
我的角色Role
负责学习闭环、开源复用取舍、数据迁移与验收边界；实现由 AI 编程工具协助完成。Organized the cards, practice flow, website integration, and Japanese playback boundaries, with implementation assisted by AI coding tools.
交付结果Outcome
v2 核心流程通过领域、迁移、浏览器、断网和自动可访问性检查；读音教学质量、真实手机/读屏和长期学习效果仍待验收。The v2 core passed domain, migration, browser, offline-reload and automated accessibility checks. Pronunciation, real-device and screen-reader acceptance, and long-term learning outcomes remain unvalidated.
产品洞察Product insight
发音学习先需要一个随手能打开的小工具。把声音、字形与回忆练习放在同一张卡片里，并明确语音来源和浏览器限制，比增加账号与复杂课程更适合这个原型。A small pronunciation tool should be easy to open. Put sound, writing, and recall together, and explain voice sources and browser limits before adding accounts or a course system.
项目简介Overview
声音手帖 v2 是浏览器本地的日语假名学习器，提供按行练习、选择题、罗马音输入、听音选假名、FSRS 间隔复习和学习备份。准备离线包后可断网使用；104 段既有 AI 音频保留，读音与新音色尚待验收。Sound Notebook v2 offers row selection, multiple-choice and typed romaji answers, audio-to-kana questions, FSRS reviews, and local backup import/export. A prepared offline package keeps the page and 104 existing AI clips available without a connection. Pronunciation quality and replacement voices still need review.
技术栈Tech stack
结果证据Evidence
2026-10-29 按研究报告重建 v2：104 个假名、平片字形分开调度的 208 个卡片状态。The October 29, 2026 rebuild covers 104 kana with 208 independently scheduled hiragana/katakana card states.
日语清单包含 104 段已有 WAV 文件；文件结构与路由测试不代表逐条读音或音色验收。The Japanese manifest lists 104 existing WAV files. File and routing checks do not establish pronunciation or voice acceptance.
使用 ts-fsrs 5.4.2 适配现有 Node22 发布副本；由 Workbox 生成仅作用于学习器路径的离线缓存。ts-fsrs 5.4.2 preserves the existing Node22 deployment contract; Workbox generates a worker scoped to the learning tool.
旧版数据先备份再迁移；导入验证结构与大小并保留替换前备份，不提供账号或自动跨设备同步。Legacy data is retained and backed up before migration. Backup imports validate structure and size and preserve the previous local state.
项目截图Screenshots
关键决策Key decisions
答对一次与真正记住并不等同。One correct answer does not establish durable memory.
将答案正确率与显式记忆评价分开，由成熟 FSRS 库安排复习；旧自标记不伪造历史复习。Separate answer accuracy from explicit memory ratings and use FSRS for scheduling; do not invent review history from old self-marked cards.
文件接入可能被误写成读音已经正确。File integration could be mistaken for pronunciation validation.
只读取已有音频清单，分别记录文件验证与待试听项；回退语音必须使用日语音色。Distinguish imported files from listening acceptance, and only use Japanese voices for playback fallback.
想看完整的代码和实现细节？Want the full code and implementation details?
在 GitHub 上查看View on GitHub