Changelog

One-cell widget: the icon now fills its cell at full size without being cut off at the corner, on any launcher.

Smaller icon on the one-cell widget; the whole cell still responds to a tap.

Widget refresh.

• Shrink the widget to one cell and it becomes just the VoiceTodo icon. Tap it to start talking. The icon is white on dark wallpapers and dark on light ones.
• At wider sizes it stays the input bar, now with the VoiceTodo icon in place of the mic.

Cleaner widget: the app icon is gone, leaving just the input bar and the mic.

Add tasks from the home screen without opening the app.

• Tapping the widget now opens a small panel over your home screen instead of the app.
• Tap the bar to type, or hold the panel's field to talk, just like in the app. Tap the mic to start talking right away; it stops when you finish.
• The panel shows what was understood, coloured by topic. Tap ✓ to save and you're back on the home screen. ✏ opens the app for detailed edits, ✕ cancels.
• Holding the widget itself still moves it; Android reserves long-press on widgets for that.

Home-screen widget, and tasks coloured by topic.

• Widget: add VoiceTodo's bar to your home screen. Tap it to type a task; tap the mic to talk. Talking from the widget stops by itself when you finish, or tap ■.
• Topics: under menu → Topics, set up topics with a name, a colour and keywords. Tasks are coloured automatically by the topic whose keywords they mention. Work, Study and Life are there to start with; edit or delete them freely.
• Tap a task to pick its topic by hand.
• Topics only colour tasks; everything stays in one list.

New app icon: a microphone that is also a pen.

More speech packs, and much better mixed Chinese-English recognition.

• New speech pack: FunASR Nano (recommended). It handles Chinese with English words mixed in ("周六去参加一个lecture", "下周一要交essay", "帮我book一个meeting room") far better than SenseVoice, and copes better with accents.
• New speech pack: Zipformer, a small and fast option (72 MB).
• Settings → Speech pack now lists all three. Download, pause or delete each one, and pick the one in use with the radio button.
• If you already have SenseVoice, it stays in use until you switch. Tap the download icon next to FunASR Nano, then select it.
• If the microphone records only silence, the app now says so instead of "didn't catch that".
• On phones where the default recording source isn't available, the app falls back to the regular microphone.

Downloaded from inside the app (Settings → Speech pack).

Model: sherpa-onnx-zipformer-zh-en-2023-11-22 from k2-fsa/sherpa-onnx (int8 encoder and joiner).

Downloaded from inside the app (Settings → Speech pack).

Model: Fun-ASR-Nano-2512 (FunAudioLLM), int8, converted by k2-fsa/sherpa-onnx (csukuangfj/sherpa-onnx-sense-voice-funasr-nano-int8-2025-12-17).

Fixes hold-to-talk after typing, and trims the UI down to icons.

• Fixed: after tapping the input field, holding it no longer started a recording. Tapping anywhere else or closing the keyboard now leaves the field, and hold-to-talk works again.
• Buttons are icons wherever an icon is clear enough: back, menu, send, swipe actions, play, refresh, delete, save and cancel.
• Removed text that said what the screen already shows.

VoiceTodo now speaks English and Chinese.

• The app is now called VoiceTodo, with an English and a Chinese interface. It follows the system language, or pick one under Settings → Language.
• Voice input understands both Chinese and English, including Chinese with English words mixed in.
• English tasks get their dates and times too: "call Bob tomorrow at 3pm", "submit the report by Friday", "remind me 30 minutes before".
• Fillers and corrections are cleaned up in English as well ("um", "Friday, no wait, Saturday").
• Open and completed tasks are now in one list: open tasks stand out, scheduled ones get a coloured edge, completed ones fade out.
• The title at the top of the home screen is gone.
• Proper nouns written in Latin letters now fix their own capitalization ("bob" → "Bob").

任务可以左右滑动操作，说出来的任务会保留录音。

• 向左滑：完成（在「已完成」里向左滑是恢复）
• 向右滑：删除，底部会出现「撤销」
• 去掉了勾选框和删除按钮
• 用语音录入的任务会保留录音，点 ▶ 可以回放
• 「设置 → 录音」可以一次删除所有已完成任务的录音

界面重做，安装包从 265 MB 变成 44 MB。

⚠️ 从旧版本更新后，需要到「菜单 → 设置 → 语音包」重新下载一次语音包（约 230 MB），之后的更新都只有几十 MB。

• 首页只保留任务列表和底部输入框，去掉了说明文字
• 左上角菜单按钮（或从左侧往右滑）打开菜单：专有名词、设置
• 设置：语音包下载（可暂停、继续、删除）、提醒权限、检查更新、软件说明
• 有需要处理的事项时，菜单按钮和「设置」上会有红点
• 没下载语音包时长按输入框，会直接跳到设置

Downloaded from inside the app (Settings → Speech pack); no need to download it by hand.

Model: SenseVoice int8 (2025-09-09) from k2-fsa/sherpa-onnx.

底部改成一个输入框：点一下打字，长按说话。

• 点输入框：出现光标和键盘，输入后点「发送」或键盘上的发送键
• 长按输入框：开始录音，松开识别；按住时上滑可以取消这次录音
• 录音时输入框会显示音量，确认手机在听
• 去掉了原来的大麦克风按钮和「打字」弹窗

说话时的犹豫和语气词会自动去掉。

• 去掉「嗯」「呃」，以及停顿时说的「那个」「就是说」「对了」等（「那个方案」这种有意义的不会删）
• 结巴重复会合并：「我我我」「明天明天」
• 改口会保留后说的：「明天，不对，后天」→ 后天；「下午三点不是四点」→ 下午四点
• 「明天，下午三点」这种中间停顿的时间会合成一个
• 确认卡片上会列出去掉了什么，点「撤回」可以恢复原话

新增「专有名词」：公司名、人名、地名不再被听错。

• 右上角「专有名词」里添加常用的词，识别结果中发音相同的字会自动改成你的写法（如 疼讯 → 腾讯）
• 英文名、读音对不上的词，可以填它“常被听成”什么（如 Shopee ← 虾皮）
• 这些词不会被误当成时间（“周一科技”不再被当成周一）
• 在确认卡片上手动改了名字，保存后会询问是否加入，以后自动纠正
• 自动纠正的地方会标出来，可以一键撤回
• 修复：输入文字时键盘会挡住「保存」按钮

首个版本。

• 按住说话，离线识别（SenseVoice），不联网
• 自动拆出任务和截止时间，保存前可以修改
• 到期前和到期时准时提醒，通知上可直接完成或延后 1 小时
• 应用内检查更新：每天一次，或点右上角「检查更新」

安装：下载下面的 VoiceTodo-0.1.0.apk，在手机上打开安装（需要允许「安装未知应用」）。仅支持 64 位手机（arm64）。