Realtime Translator – Live Subtitles & Voice is a Windows tool for translating audio playing on a PC into floating subtitles and optional spoken output. The official description covers meeting applications, browsers and video players, with either all system audio or one selected app as the source. It is aimed at understanding a conversation or recording across languages, not changing what another participant says.
This First look is based on the Microsoft Store description. We have not translated a meeting, tested latency or compared language accuracy. Recognition, translation and generated speech are separate stages, and the chosen configuration determines which material stays local or reaches an external provider.
Based on official store information. We have not tested this app hands-on.
What it does
The developer describes local speech recognition through Whisper and local voices through Piper, with GPU support. For translation, the choices include Anthropic Claude, an OpenAI-compatible API or local Ollama and LM Studio setups. A local recognition stage does not mean the translated text is necessarily processed locally; that depends on the selected translator.
Subtitles float above other windows, and the translation can also be read into headphones. Each output can be switched with a shortcut. The description distinguishes provisional translation while someone is speaking from a committed line after the sentence ends. That makes the evolving text a draft rather than a final statement to act on immediately.
Meeting and Movie profiles provide starting configurations. A glossary is intended to preserve names and specialist terms, and a bilingual transcript is saved on the PC. These are useful points to test with material resembling your real work. We have not verified glossary coverage or whether the saved transcript captures every interruption or speaker change.
A practical first setup would use a non-sensitive recording with a known translation. Select the intended application as the audio source, inspect the recognised words and compare both languages. Then try the spoken output through headphones. The source specifically recommends headphones to keep the generated voice from leaking into a meeting.
The tool's interface languages are listed separately from translation capability. The retained description names English, Vietnamese and Japanese for the interface but does not supply an exhaustive supported language-pair table. Confirm the pair and models you need rather than inferring coverage from a general translation claim.
Before you install
The privacy description says audio is not saved to a file, while text goes only to the translator you choose and a bilingual transcript remains locally. Keep those distinctions intact. Not recording audio is not the same as retaining no meeting content, and an external translator has its own data-handling terms.
Before using a client or workplace conversation, establish whether the participants and applicable policy permit this processing. A tool that can listen to system audio may include information beyond the sentence you intended. Choosing one application can narrow the source, but we have not audited capture boundaries or provider requests.
Automatic translation can alter names, numbers, qualifications and tone. Verify important commitments directly rather than relying on a provisional subtitle. The app can support understanding, while a consequential decision may need an interpreter or another appropriate confirmation route. We have not evaluated accuracy for legal, medical or other specialised exchanges.
Local models also have practical requirements. The source specifies Windows 11, 64-bit, and recommends a graphics card for recognition speed. We have not benchmarked supported GPUs or established resource use. A free local translator still requires suitable models, configuration and computing capacity.
For a meeting, test the full sound arrangement before others join. Generated speech routed into the wrong output can distract participants or be captured again. We have not assessed feedback prevention or every application's audio handling.
Price and access
The US Microsoft Store offer is free. The free version includes translation, subtitles and voice for sessions up to fifteen minutes, with starting a new session described as a way to continue. Pro removes the time limit and adds AI meeting summaries; the retained source does not give its price. Cloud translation uses your own API access, which may incur charges. Confirm current tier and provider terms separately.
Our take
The strongest idea is making the processing route visible: choose what to hear, which translator handles the text and whether subtitles or voice are useful. That flexibility can fit different privacy and equipment needs, provided the setup is understood before a live conversation.
Begin with known material and inspect recognition as well as the final translation. Check transcript retention, the fifteen-minute free boundary and the actual audio output. If the result is uncertain, preserve that uncertainty instead of letting a fluent generated voice make it sound settled.
We have not validated latency, accuracy or privacy implementation. Our take is a source-based introduction to a configurable translation workflow, with participant permission, provider choice and verification of important meaning central to its practical use.
For a glossary test, include a name or technical term whose spelling matters. Check both the subtitle and saved transcript. A correct-looking word in one output does not establish that every stage used the same interpretation.
Sources
App icon from the official store listing.
