Changelog

New: Screen tab (beta). Select text on your main screen - game subtitles, chat, quests - and SyncLingo translates it as it changes, in its own block or on top of the original. Recognition runs locally on Windows OCR (needs the Windows language pack); translation uses your Incoming provider. In games, use borderless windowed mode.

New: file transcription. The Utilities tab turns an audio or video file (mp4, m4a, mp3, wav, flac, ogg Vorbis) into a .txt or .srt.

Also new - Proxy support (http:// or socks5://, username and password supported) in Settings → API. - Local-only mode in Settings → About: no cloud speech recognition or cloud voices. Translation still needs the internet. - The API tab shows what DeepL, ElevenLabs and Deepgram have left, where your key allows it. - AssemblyAI: universal-3-6-pro adds real-time Russian, Korean and Romanian. Real-time 3.x Pro models now receive the source language, which improves recognition. Per-phrase universal-3-5-pro now answers in under 0.2 s (median, our tests). - DeepL receives the previous phrase as context. - Speaking speed for the Windows voice, and a run timer on the overlay.

Faster - New installs get shorter phrases by default, so subtitles arrive about 1.5 s sooner. Updating? Lower "Max recording length" on the Incoming or Outgoing tab for the same effect. - Local Whisper uses more CPU threads on PCs with 10+ logical processors.

Clearer errors - In per-phrase mode, a rejected key stops the direction with one message instead of an error per phrase. - A lost connection reads as "no connection" in your language and clears once phrases flow again. - Warnings when the paid translator falls back to Google, when real-time recognition keeps failing to reconnect, and when the app has to skip phrases. - A blocked microphone links to the Windows privacy page. - A failed model-list request shows its real cause instead of "No STT model available".

Fixes - A real-time session that stays connected but stops answering is now detected and reconnected. - Whisper downloads resume after an interruption and are checksum-verified. - "Automatic" interface language now follows the Windows display language. - Voice lists for providers with 50+ voices; a Ukrainian spelling fix.

SHA-256: 519bf401e502947ee5a9d208cc9679a08f389efa3beb13dcf1674861bef85d3a VirusTotal: https://www.virustotal.com/gui/file/519bf401e502947ee5a9d208cc9679a08f389efa3beb13dcf1674861bef85d3a

A big one — a new look, a new icon, and a good bunch more languages.

Eight interface languages. Until now the app only spoke English — it now also comes in Russian, German, Spanish, French, Brazilian Portuguese, Ukrainian and Simplified Chinese. It picks up whatever your system is set to on its own — or you can just choose one yourself in Settings → About.

A fresh look. We reworked the whole interface around one simple idea: green is what's coming in — the speech we decode for you; amber is what's going out — your voice on its way to everyone else. Panels are cleaner, the little readouts are easier to catch at a glance, and it finally matches the website.

A new icon. Two chevrons pointing in and out — you'll spot it on the window, in the taskbar, and in the installer.