IN ONE SENTENCE

重点不是得到原始转写，而是得到能直接发出去的文字。

轻语输入适合聊天、写作、工作说明、任务拆解和中文口述后输出英文。它会去掉明显口头语、整理标点和层级，并尽量把结果插入光标位置；插入失败时自动复制到剪贴板。

它不会替你编造没有说过的背景。目标是忠实整理表达，而不是把语音变成新的聊天机器人。

IN ONE SENTENCE

The goal is not raw transcription. It is text ready to send.

Whisper Input fits chat, writing, work instructions, task breakdowns, and English output from Chinese speech. It removes obvious filler, improves punctuation and structure, then inserts the result at the cursor or copies it to the clipboard as a fallback.

It does not invent context the user never said. The goal is faithful cleanup, not a new chatbot.

INPUT FLOW

不切换窗口，也不打断正在写的东西。

1. 按下全局快捷键在聊天框、文档、Issue 或任何当前输入位置开始口述。
2. 发送到已配置的云端 ASR识别中文与工作场景中的英文术语。
3. 按选择的风格整理原文、轻度润色、清晰结构或正式表达。
4. 插入当前光标直接回到正在写的地方，不需要手动复制。
5. 失败时复制到剪贴板即使目标应用不允许自动插入，也能拿到结果。
6. 按设置保存历史在本机回看、复制或删除最近输入。

INPUT FLOW

Stay in the current app and keep writing.

1. Press the global shortcutStart speaking in a chat field, document, issue, or any active input.
2. Send audio to the configured cloud ASRRecognize Chinese speech and mixed English terms.
3. Apply the selected styleRaw, light polish, clear structure, or formal expression.
4. Insert at the current cursorReturn text directly to the place where the user was writing.
5. Fall back to clipboard copyKeep the result even when the target app blocks insertion.
6. Save local history if enabledReview, copy, or delete recent inputs on the device.

REAL INTERFACE

四张真实界面，分别回答状态、风格、模型和隐私。

这些图片来自当前 GitHub README，不是重新绘制的概念稿。

REAL INTERFACE

Four real screens for status, output styles, models, and privacy.

Every image comes from the current GitHub README rather than a recreated concept mockup.

CORE CAPABILITIES

从说话到上屏，中间每一步都服务于“可用”。

CORE CAPABILITIES

Every step between speech and insertion is designed for usable output.

FOUR OUTPUT STYLES

同一段口述，按用途整理成四种结果。

原始口述

第一把代码推上去，第二改 README，第三发安装包。

清晰结构

1. 推送代码
2. 更新 README
3. 发布安装包

FOUR OUTPUT STYLES

One spoken thought, organized for four different uses.

Spoken input

First push the code, second update the README, third publish the installer.

Clear structure

1. Push the code
2. Update the README
3. Publish the installer

DATA AND PRIVACY

这是 cloud-first 产品，设置页会明确告诉你数据去了哪里。

当前主线不是离线 ASR。使用前需要按界面配置语音识别和文字润色服务，并理解相应服务商的数据政策与费用。

DATA AND PRIVACY

This is cloud-first, and the settings page makes every data destination explicit.

The current product is not offline-first ASR. Users configure speech and language services and remain responsible for provider policies and cost.

INSTALL AND USE

安装后只需要完成模型设置，就可以在任意输入位置开始说话。

1. 从 GitHub Releases 下载最新版 Windows 安装包。
2. 安装并启动轻语输入，进入“设置 - 模型设置”。
3. 选择千问或豆包等 ASR，并配置一个文字润色模型。
4. 设置全局快捷键和输出风格。
5. 回到任意输入框，按快捷键开始口述。

INSTALL AND USE

After model setup, start speaking from any input field.

1. Download the latest Windows installer from GitHub Releases.
2. Install and open Whisper Input, then visit Model Settings.
3. Choose an ASR provider and configure a model for text cleanup.
4. Set the global shortcut and output style.
5. Return to any input field and press the shortcut to speak.

PRODUCT BOUNDARIES

保持输入工具的定位，刻意不把它做成一套大平台。

• 不注册 Windows 系统输入法，不接管系统 IME。
• 不做会议记录平台，专注短到中长文本输入。
• 不主动生成用户没有说过的信息。
• 不加入 RAG 或 Agent 工作流，保持输入链路轻量。
• 当前主线以真实可用的云端 ASR 为主，不宣传离线优先。

PRODUCT BOUNDARIES

Stay focused as an input tool instead of becoming a large platform.

• It does not register as a Windows system IME.
• It is not a meeting-recording platform.
• It does not invent information the user did not say.
• It does not add RAG or agent workflows to the input path.
• The current product uses practical cloud ASR rather than claiming offline-first operation.