LiveCaptions-Translator adds a translation API behind Windows captions

The recogniser belongs to Windows, not to this project

The architecture is stated as an equation in the README: Windows LiveCaptions plus a Translate API. Everything this tool does sits after recognition. That is why the description emphasises that it is lightweight, and why one specific machine class is explicitly not required, since it enables real-time speech translation without needing a Copilot+ PC. The reasoning given for building on the operating system component is stated in three claims: the built-in feature is easy to use, uses few resources, and has extremely high recognition accuracy. So the accuracy question is answered by not doing the recognition at all, and the tool's own scope is the translation stage and the interface around it. Distribution is a single click from the releases page, with no installer described.

Nine engines, three of which can be self-hosted

The engine table is the core of the configuration surface, and it is organised by type and hosting rather than by quality. The LLM-based entries are Ollama, an OpenAI compatible API and OpenRouter. The traditional entries are Google Translate, DeepL, Youdao, Baidu Translate, MTranServer and LibreTranslate. Hosting is listed per row, and three of the nine are self-hosted: Ollama among the language models, and MTranServer and LibreTranslate among the traditional translators. The rest are online services. The feature list also states that two engines work out of the box, naming Google Translate among them, and that language model engines are strongly recommended, on the stated grounds that they handle incomplete sentences and understand context, which is exactly the failure mode of subtitle translation.

The overlay window is what makes it usable for streams and games

The feature that distinguishes the tool from a plain translation panel is a borderless, transparent overlay window for displaying subtitles, described as the most immersive option and specifically useful for gaming, video and live streams. It can be made completely embedded into the screen so that it becomes part of the desktop and does not interfere with other operations, which the README frames as ideal for gamers. The overlay is opened from the taskbar and its parameters are exposed: window background, subtitle colour, font size and transparency, with the description that the configurability is high enough to match a preference exactly. There is also a count to set, in the Overlay Sentences section of the settings, which decides how many sentences are displayed at once. So the settings page is where the difference between a translation window and a subtitle layer is actually decided.

LiveCaptions is hidden after the first run and has to be reconfigured by hand

The integration claim is that the tool invokes Windows LiveCaptions automatically without opening a separate window, and that after the first use LiveCaptions itself is hidden by default, with a button available in the settings to show it again. Two manual steps sit underneath that automation. The first is enabling the option to include microphone audio in the LiveCaptions settings, which is what makes microphone speech reach the translation stage at all. The second is marked important twice in the file: the source language has to be changed inside Windows LiveCaptions. So the tool does not choose the input language for you, and a run that translates nothing is usually a configuration problem in the operating system rather than in the application.

The floor is Windows 11 22H2 with a .NET 8 runtime

The prerequisites are short and unforgiving. Because the tool is built on LiveCaptions, the requirement is a Windows version with LiveCaptions support, which has been available since Windows 11 22H2. The requirements table carries a second row recommending a configuration and noting that earlier versions were not tested. The runtime recommendation is .NET runtime 8.0 or higher, and there is a documented fallback for people who cannot install one: a build distributed with the runtime included, at the cost of a larger download. There is no macOS or Linux path in the prerequisites, and no mention of a Windows 10 build that would work, which matches the fact that the platform feature the tool depends on arrived in Windows 11.

Getting started is five ways to confirm the caption feature exists

The first run is deliberately about verification rather than configuration. Before the tool will do anything, you have to confirm that Live Captions is available on your system, and the README lists five routes for doing that: toggling Live captions in quick settings, pressing the Win, Ctrl and L key combination together, going through Quick settings and then Accessibility to Live captions, starting from the Start menu through All apps to Accessibility, and using Settings, Accessibility, Captions to enable Live captions. The second step is configuring LiveCaptions itself, where Windows asks for consent to process audio on first start, and the file links to Microsoft's own guide on using live captions for the detail. The tool's own wiki is linked from the same area for anything beyond that.

History and log cards are two different retention models

Two features look similar and behave differently. History Management records both the original and the translated text, aimed at meetings, lectures and important discussions, and every record can be exported as a CSV file, which is the format to reach for when you need the transcript outside the application. Log Cards are a display feature rather than a storage one: recent transcription records are shown as cards so that the surrounding context is easier to grasp, they can be enabled from the taskbar of the main page, and their number is configurable in the Log Cards section of the settings. So the transcript is kept in full and exported on demand, while the cards are a small adjustable window onto the most recent entries. Along the same lines, the interface offers always on top, pause and resume for translation, and copying text with a single click.

Version numbers carry a build counter and releases stopped in January

The release tags are four segment numbers rather than the usual three: v1.6.1253.1718, v1.6.1254.2719 and v1.7.1300.1822, with the second and third segments behaving like build counters that increase with each release rather than like a semantic version. The dates show an irregular cadence, with the middle tag published on 2025-09-27, the first on 2025-06-17 and the newest on 2026-01-18. The last push to the default branch master is dated 2026-08-17 and the repository is not archived, so work has continued on the branch without a release tag for several months. The tree is a small C# solution: a project file and a solution file for the application, a src directory, a Properties directory, images, and two READMEs, one English and one in Simplified Chinese. The licence is Apache-2.0 and continuous integration runs through a dotnet build workflow.

Editorial conclusion

This tool suits someone who already relies on Windows LiveCaptions for transcription and wants the same stream translated, especially for meetings, lectures or live streams where a floating subtitle overlay is the point. It does not suit a Windows 10 machine, since the caption feature it builds on arrived in Windows 11 22H2, and it does not suit someone who needs offline speech recognition, because the recogniser is not part of this project. Before relying on it, pick an engine on privacy grounds, since three of the nine can be self-hosted, and note the release tags stopped in January 2026 while the branch has moved since.

Frequently asked questions

Official sources

1. Issues
2. License: Apache-2.0
3. README
4. Releases
5. SakiRinn/LiveCaptions-Translator on GitHub