v0.6.0

What's Changed

• feat: unified branding, in-app auto-updater, English UI, and UX polish by @NBS282 in #19

Full Changelog: v0.5.1...v0.6.0

v0.5.1

What's Changed

• fix(setup): stop the setup UI freezing during large model downloads by @NBS282 in #18

Full Changelog: v0.5.0...v0.5.1

v0.5.0

What's Changed

• feat: native STT/AST via transcribe.cpp (GGUF) — RAM cut and cross-platform core by @NBS282 in #17

Full Changelog: v0.4.0...v0.5.0

LiveTranslate v0.4.0 — Faster setup & new translation engine

Highlights

New translation engine (default)

• Parakeet TDT 0.6B v3 + MarianMT cascade replaces Canary 1B Flash as the default engine: faster decoding on CPU, ~2 GB less RAM, and no more repetition loops on short phrases ("no, no, sí, claro" now translates correctly).
• The source-language transcript is back in the subtitles (ES: … → EN: …).
• Canary 1B Flash remains available as an alternative via LT_TRANSLATION_ENGINE=canary (its model downloads automatically on first use).

Selectable language pair

• Choose the translation direction in the live session: EN ↔ ES / FR / DE (6 pairs).
• The pair is validated end-to-end and locked while a session is running.

Faster first-run setup

• First-time download halved: ~3.6 GB (Canary is no longer pre-downloaded).
• Downloads now retry with backoff, resume interrupted transfers, and repair truncated files automatically.
• Models download two at a time, with a real progress bar driven by actual cache growth.

Faster app startup

• The engine binds its port immediately and loads models in parallel in the background.
• The UI shows real warmup progress ("Loading translation models… 66%") instead of a frozen message.
• A fatal model failure is reported instantly instead of after a 10-minute timeout.

Audio fixes

• Translated audio is no longer clipped at the end: playback now waits for the device to actually finish rendering (plus a silence tail for device latency).
• Cloned-voice synthesis no longer drops quiet words — the silence trim threshold is now relative to each clip's own level (fixes "Yes, of course." playing as just "Yes").

Full Changelog: v0.3.0...v0.4.0

LiveTranslate v0.3.0 - Simultaneous translation

The interpreter update: LiveTranslate now translates while you speak.

What's new

• Simultaneous translation - sentences are detected as you talk and voiced while you keep speaking, interpreter-style. No more waiting for you to stop.
• Live streaming subtitles - the in-progress translation appears and refines in real time, in the main window and the floating overlay.
• New speech-translation engine - NVIDIA Canary 1B Flash translates Spanish speech directly to English text in one pass (the previous two-stage pipeline remains available as a fallback via LT_TRANSLATION_ENGINE=legacy).
• Sturdier under long speech - every segment is translated in order (nothing is dropped), quiet microphones are normalized, and segments the decoder rejects are recovered by re-decoding in halves.
• Diagnostics - segment lifecycle logging and bounded retention of failed-segment audio for troubleshooting.

Notes

• First run downloads ~5.6 GB of models (one-time; adds Canary 1B Flash).
• ~8 GB RAM recommended.

LiveTranslate v0.2.1 - Voice Cloning fixes

Fixes the voice cloning failures reported in v0.2.0:

• The engine no longer dies at startup when the cloned-voice model cannot load; the app degrades to the standard voice and reports why.
• Recording is blocked until the engine is ready, so recordings are never lost to a startup race.
• Saving a voice profile no longer causes translation errors (500) until restart: the voice state export is now atomic and cloned-synthesis failures fall back to the standard voice per phrase.
• Engine logs are now written to %LOCALAPPDATA%\com.livetranslate.app\logs\engine.log for diagnosis.

Note: v0.2.0 installers stop working for new installs (expired model-download credentials). Use v0.2.1.

LiveTranslate v0.2.0 — Voice Cloning

Voice cloning: hear the translation in your own voice (Pocket TTS).

LiveTranslate v0.1.0

First public release

Real-time speech translation with transparent on-screen subtitles — 100% local, no cloud, no API keys.

Download

File	Description
LiveTranslate_0.1.0_x64-setup.exe	NSIS installer (recommended)
LiveTranslate_0.1.0_x64_en-US.msi	MSI installer

Requirements

• Windows 10 or 11 (64-bit)
• ~4 GB free disk space
• Internet connection on first run (models download once, then fully offline)

What's included

• Spanish → English real-time translation
• Transparent always-on-top subtitle overlay
• Microphone and system audio capture (WASAPI loopback)
• Push-to-talk mode with configurable hotkey
• One-click setup: downloads Python runtime + AI models automatically
• Works completely offline after initial setup

Known limitations

• Windows only (Linux/macOS planned)
• Spanish → English only (multilingual support planned via NLLB-200)