VoxCast
100% 本地 · 在你的 GPU 上离线运行100% 本地 · 离线 · GPU

开口说。 不用打字。

按下热键，开口说话，文字直接落进当前窗口。语音识别完全在你的电脑上运行：没有云端，没有上传，没有订阅。

免费下载 工作原理
就绪

点击麦克风开口说：文字随即出现。

100% 本地 免费 Windows 10 & 11 热键 Ctrl Shift Space

工作原理

从想法到文字，只需三步。

没有繁琐配置，无需注册云端账号。安装，按下热键，开干。

按下热键

随时随地开始

按 Ctrl+Shift+Space，不管当前打开的是什么程序。VoxCast 都在听。

Ctrl + Shift + Space
开口说话

说就行了

说出你想写的内容。面板会实时显示正在录音。

文字出现

直接进入窗口

你说的内容立刻出现在当前窗口，通过键入、剪贴板或两者兼用。

功能

全部本地。全归你。

听写、转换、润色。每项功能都在你的硬件上运行。

全局热键听写

一个快捷键就能启动和停止录音。识别出的文字直接进入当前窗口，无论是浏览器、邮件客户端还是编辑器。不用切换窗口，不用手动粘贴。

录制会议

系统音频加麦克风：VoxCast 可录制 Zoom、Meet、Teams、WhatsApp 及任何其他应用，并在本地生成带时间戳的转写稿。

AI 润色

把你的听写变成一封成稿邮件或利落的文字。可选功能，使用你自己的 Anthropic 或 OpenAI 密钥。

文件 → Markdown

音频和文档在本地变成整洁的 Markdown，右键或拖放即可。保存位置由你决定。

GPU 加速

完整的 CUDA 运行时已打包进安装程序，即使没有安装 CUDA 工具包也能运行。没有 GPU？VoxCast 会改用 CPU。

三档精度可选

Swift、Precise 或 Ultra：从轻若鸿毛到近乎零错，取决于你的电脑能跑多快。

多语言

用中文、英文及更多语言听写。VoxCast 识别范围极广，无需在脑中来回切换。

自动更新

新版本一发布，VoxCast 就会通知你。什么都不用手动检查。

会议

每场会议都变成文字。

VoxCast 同时录制系统音频和你的麦克风，无论是 Zoom、Google Meet、Teams、WhatsApp 通话还是浏览器。你也可以只录一个应用：音乐、系统提示音和其他一切都完全不会进入转写稿。哪怕是数小时的会议，也会以音频文件加带时间戳的转写稿保存到你的硬盘上，完全本地。还可以选择让 AI 润色在开头附上一份包含决议和待办事项的摘要。

适用于

ZoomGoogle MeetTeamsWhatsApp+ 任何应用

守规矩：录音前请告知参会者。在许多国家，未经同意的录音属于刑事犯罪。VoxCast 会在你首次录音前提醒你。

文件 → Markdown

文件变成 Markdown，就在本地。

把文件拖到面板上，或在右键菜单里选「转换为 Markdown」。音频由 VoxCast 在你的电脑上转写，文档由 MarkItDown 干净地解析。存到哪里，由你决定。

音频

MP3OGGM4AWAVFLAC

文档

PDFDOCXPPTXXLSXCSV文档
VoxCast Server

一台强力 PC。整个网络的转写。

VoxCast Server 在一台配备强力 GPU 的机器上运行语音识别，网络里的每个 VoxCast 都能使用它：听写、会议和文件转换。你的音频不出大楼，完全不经过云端。

  1. 1在配备强力 GPU 的机器上安装 VoxCast Server。安装程序会自动注册 Windows 服务并放行防火墙。
  2. 2在任意 PC 上打开 VoxCast 设置，启用远程服务器，点击“查找服务器”。一键即可填入地址。
  3. 3粘贴 API 密钥（开始菜单：“VoxCast Server API key”），保存即可。此后所有转写都在服务器上运行。
Windows 版服务器 安装指南

Linux 或 Docker？下载服务器安装包，按照安装指南操作即可。

隐私

什么都不会 离开你的电脑。

你的录音只在本地处理，就这么简单。没有上传，没有云服务，也没有任何可能存下你声音的账号。语音模型存在你的硬盘上，AI 润色用的 API 密钥安全地保存在 Windows 凭据管理器中。对于涉密听写，这是兼顾隐私与 GDPR 合规的选择。

  • 无云端转写音频和文字都留在你的硬件上。
  • 无账号，无订阅语音识别分文不取。
  • 这个网站也说到做到无 Cookie，无跟踪器，字体自托管。
你的数据

一切本地。润色只在你需要时。

你的数据就这样流动：到此为止，一步都不会多走。

橙色发光的音频波形，完整封闭在带绿色状态点的发光边框内：你的信号不出设备
始终 · 100% 本地

听写、会议、文件

录音、语音识别和转写稿全部在你的 GPU 或 CPU 上完成。任何数据都不会离开你的电脑：没有上传，没有账号，没有云端。

  • 完全离线可用
  • 音频 & 文字都留在你的硬盘上
  • 无订阅，无隐藏费用
紫色音频波形穿过一个发光的光之钥匙孔：只有你的 API 密钥才能打开通路
可选 · 仅用你的密钥

AI 润色

只有当你主动开启时，成稿文字（绝不是你的音频）才会用你自己的 Anthropic 或 OpenAI API 密钥进行润色。密钥在你手里，决定权也在你手里。

  • 默认关闭，直到你启用
  • 只发送文字，绝不发送音频
  • 你的密钥，安全存放在 Windows 凭据管理器中

没有密钥，就没有对外连接。VoxCast 照样全力听写、转写和转换。

为谁而做

为所有写得多的人而生。

重度写作者

邮件、笔记和文字，说出来比打出来快。

开发者

听写注释、提交信息和文档，双手不离键盘。

记者 & 作者

把采访和语音备忘变成可搜索的本地文字。

无障碍

用声音操作电脑，在任何程序里，不依赖云端。

全面掌控

按你的工作方式来设置。

麦克风、语音模型、语言、输出方式和热键：全都集中在一处，和面板一样的深色外观。在 Swift、Precise 和 Ultra 之间选择，剪贴板或键入随你，保存位置也由你定。

模型：Swift / Precise / Ultra 输出：剪贴板 / 键入 / 两者 语言：自由选择
深色的 VoxCast 设置窗口，包含音频、识别和输出区域，带有选择框和开关。
FAQ

常见问题。

VoxCast 免费吗？
免费。VoxCast 是 Windows 10 和 11 上的免费下载。语音识别本身没有任何订阅，它完全在你的电脑上运行。只有可选的 AI 润色会用到外部服务商和你自己的 API 密钥。
语音识别真的能离线运行吗？
是的。转写 100% 在你自己的 GPU 或 CPU 上本地完成。模型下载完成后不再需要联网，也不会有任何音频上传到云端。
支持哪些文件格式？
音频支持 MP3、OGG、M4A、WAV 和 FLAC。文档支持 PDF、DOCX、PPTX、XLSX、CSV 以及图片。全部在本地处理成整洁的 Markdown。
我需要 GPU 吗？
不需要。VoxCast 也能在 CPU 上运行。配合兼容的 NVIDIA GPU，得益于内置的 CUDA 运行时，识别速度会快得多。无需单独安装 CUDA。
VoxCast 支持哪些语言？
中文、英文及更多语言。VoxCast 覆盖约 99 种语言。所有支持的语言都可以完全离线进行语音识别。
我的音频会离开电脑吗？
不会。录音和转写完全留在本地。只有当你启用可选的 AI 润色时，生成的文字才会发送给你选择的服务商（Anthropic 或 OpenAI）。
AI 润色是怎么工作的？
可选功能，使用你自己的 Anthropic 或 OpenAI API 密钥。你的听写会被改写成邮件、回复或利落的文字。你可以创建自己的模式；密钥安全地存放在 Windows 凭据管理器中。
VoxCast 能转写会议和视频通话吗？
能。VoxCast 会录制系统音频，可选加上你的麦克风。Zoom、Google Meet、Teams、WhatsApp Desktop 及任何其他应用都适用，无需任何插件。你还可以只录制单个应用，音乐和系统提示音就完全不会混入。哪怕数小时的会议也会以音频文件形式保存在本地，并处理成带时间戳的转写稿。重要提示：录音前请告知参会者；在许多国家，未经同意的录音属于刑事犯罪。
系统要求是什么？
Windows 10 或 11（64 位）。GPU 能加快识别速度，但不是必需。当前版本：1.9.1。

用你的声音写作。

下载 VoxCast，向任何窗口听写。免费、本地，几分钟即可上手。

Windows 版下载

Windows 10/11 · 推荐 NVIDIA GPU（否则使用 CPU）

选择语言

DeutschEnglishEspañolFrançaisItalianoPortuguêsNederlandsPolskiTürkçeРусский中文日本語한국어

新变化

VoxCast 每个版本一目了然。

v1.9.1最新

改进

  • VoxCast keeps working when the server is not available: VoxCast now checks at every start, and regularly while it runs, whether your VoxCast Server can be reached. If it cannot, dictation, meetings and file conversion switch to this PC (GPU or CPU) automatically, and you get a notification saying so.
  • Automatic return to the server: VoxCast keeps trying the server in the background and notifies you as soon as it is reachable again, then puts the work back on the server.
  • A recording is no longer lost when the server drops out mid-dictation: the recording is transcribed on this PC instead of ending with an error.
  • Nothing in progress is interrupted: a running dictation, meeting or live session always finishes; the switch only applies to the next one.
  • Server status in the settings: the remote server section shows whether the server is reachable and where transcription is running right now.
v1.9.0

新功能

  • VoxCast Server, transcription for your whole network: run the new self-hostable VoxCast Server on one machine with a strong GPU and every VoxCast on your network uses it for dictation, meetings, live captions and file conversion. Your audio stays on your network, no cloud involved. Free download for Windows (Linux and Docker packages included) at getvoxcast.com/server.
  • Find your server with one click: Settings, Remote server, "Find server" lists every VoxCast Server on your network; one click fills in the address, no IP addresses to type.
  • Central AI enhancement: the server can hold one company AI key, so clients enhance dictations and meeting summaries through the server without entering a key of their own. Choose "Use the server" under AI enhancement.
  • Company AI modes: define enhancement prompts once on the server and everyone can pick them, alongside their own modes.
  • Encrypted connection (HTTPS): the server can serve HTTPS with a self-signed certificate; the client trusts it once by confirming its fingerprint, then remembers it.
  • Enterprise rollout: companies can ship VoxCast pre-configured and lock the server and key settings via a managed policy file.

改进

  • Activation window behaves like a normal window: it can be moved (drag the title), shows a taskbar entry and no longer stays on top of every other window.
  • When a server is connected, the local model choice is disabled with a note, since recognition runs on the server.
v1.7.2

新功能

  • Automatic language detection for audio files: converting audio to Markdown now detects the spoken language by itself, even when it changes mid-file; before, the dictation language was forced. A new "Audio language" picker in Settings under File Conversion can still pin a language.

修复

  • Conversion always delivers a result: converting a file right after the app starts now waits for the speech engine instead of instantly claiming success with an empty or missing file; real problems show an honest error and no file is written.
  • Background jobs can no longer vanish: conversions, updates, activation and meeting actions are now protected against being cancelled silently by the runtime; a started conversion always ends with a file or a clear error.
  • First recording after launch: the waveform now reacts from the first second on every system; a slow one-time component load could freeze it during the very first recording.
v1.7.1

改进

  • Model switch with confirmation: picking a model tier that is not installed yet now asks before downloading and shows the download size; the progress window no longer asks for your dictation language again.
  • Crash reports: in the rare case of a crash, VoxCast now writes a report file that makes the cause diagnosable for support.

修复

  • First dictation after launch: a recording started right after the app launches now begins reliably and the waveform reacts immediately; it could stay frozen for around ten seconds or the recording silently never started.
  • Stability when docked: fixed a crash that could occur when the panel was docked to a screen edge and the small panel was switched on.
  • Small panel: the AI button next to the microphone is no longer cut off.
  • Smoother waveform: the level display no longer stutters on slower machines while the speech engine is busy.
  • Microphone level: the automatic input level boost before a recording works again on all systems.
v1.7.0

新功能

  • Hardware check on first start: VoxCast now detects your graphics card during setup and preselects the best model tier for your PC (Ultra on capable cards). You can change it in Settings at any time.
  • Digitally signed: the app and its installers are now signed by Hashfox GmbH, so Windows can verify the publisher.
  • Much smaller updates: starting from this version, future updates download only the changed part of the app (about 200 MB) instead of the full package.
  • Movable windows: the download and update windows can now be dragged aside; they open centered again on the next start.

改进

  • Clearer Settings: switches sit right-aligned so labels are never cut off, help texts are easier to read, and automatic language detection is now its own switch that locks the language picker while active.
  • With live transcription enabled, the text output options are greyed out and a note explains why they do not apply there.
  • The AI menu opens as a seamless extension of the panel and stays fully visible at every screen edge.
  • In the small panel, clicking the waveform stops the recording.
  • The installer now speaks 12 languages, and the Settings window gets its own taskbar button while it is open.
  • The panel shows AI enhancement in blue and previews the result in a single line.

修复

  • The microphone test no longer freezes the app on the first click; stopping early keeps the partial recording for playback.
  • The language list during setup is no longer cut off at the window edges.
  • English now shows its flag everywhere in the app.
v1.6.1

改进

  • Reliable update notifications: VoxCast now keeps checking for new versions while it runs (every few hours), not just once at launch. If the check fails because your network was still waking up, it retries within a minute.
  • Checking for updates manually in Settings now also opens the update window with the full changelog.
  • The update window waits for the right moment: it never interrupts you while dictating, transcribing or recording a meeting and appears once VoxCast is idle again.
v1.6.0

新功能

  • Live transcription: turn it on in Settings and your words appear at the cursor while you speak, instead of after you stop. Best with a graphics card; it also runs on the processor with a heads-up that it may lag.
  • Three new AI modes: To-do list, Notes and Formal join Clean up, E-Mail and Reply. Existing installations receive them automatically.
  • 13 interface languages: the app now speaks English, German, Spanish, French, Italian, Portuguese, Dutch, Polish, Turkish, Russian, Chinese, Japanese and Korean. VoxCast follows your Windows display language automatically; you can pick one in Settings, switching applies instantly.

改进

  • AI enhancement now reliably answers in the language you dictated: VoxCast tells the AI which language it heard instead of letting it guess. This also protects your own custom modes.
  • Smarter AI model selection: if the recommended model is not available for your API key, VoxCast automatically finds a suitable one (Anthropic and OpenAI).
  • Sharper built-in AI prompts: e-mails no longer invent names and keep the formality you dictated; all modes resolve self-corrections while speaking.
  • Built-in AI mode names now follow the interface language.

修复

  • Starting a dictation while VoxCast is still starting up no longer hides the waveform for that first recording.

提示

  • Live transcription types raw text at the cursor and does not use AI enhancement (that step rewrites the whole text, which streaming cannot do).
v1.5.1

新功能

  • Update dialog with changelog: when a new version is available, VoxCast now shows what's new and asks you to confirm before downloading and installing.

改进

  • The update download window is clearer, with the full change list and progress.
v1.5.0

新功能

  • Record a single app: capture just Zoom, Teams, WhatsApp, Meet or any other app; music, notifications and other system sounds stay out of the transcript.
  • Automatic meeting language: meetings are transcribed in whatever language is spoken, and handle a language switch in the middle of a call. You can still pin a fixed language in Settings if you prefer.

改进

  • Better microphone and headset compatibility when recording meetings.
v1.4.1

新功能

  • Meeting recording polished: clickable stop, notifications for every result, and clicking the "meeting transcribed" toast opens the transcript.

修复

  • USB headsets no longer prevent meeting recording from starting.
  • Unfinished meeting recordings (e.g. after a crash or quit) are transcribed automatically on the next start; nothing is lost.
  • Long dictations are no longer cut off when enhanced with AI; the original text is always recoverable.
v1.4.0

新功能

  • Record & transcribe meetings and calls: system audio plus your microphone, works with Zoom, Meet, Teams, WhatsApp and any other app. Saves an audio file and a timestamped transcript, fully local, with an optional AI summary.
  • New "Ultra" accuracy tier for the best transcription quality, and a hint in Settings showing whether a model fits your graphics card.
v1.3.5

新功能

  • Bilingual interface (English / German) with instant switching in Settings.
v1.3.4

改进

  • More robust activation and updates on unstable connections.
v1.3.3

修复

  • Fixed the "license invalid" problem: activation, updates and AI enhancement now reliably reach the server.

改进

  • Update check now runs before the activation screen, so a stuck activation can always be resolved by installing the newest version.

免费获取 VoxCast

留下你的邮箱，获取激活密钥。然后即可下载 Windows 版 VoxCast。

找回密钥

输入你获取密钥时使用的邮箱地址，密钥会直接在这里再次显示。

已购买许可证？请发邮件至 info@hashfox.com，我们会为你处理。