Beautiful Voice 0.3.0

Already on 0.2.0? The app offers this update by itself: the card in the sidebar, or Settings → About → Check for updates.

What's new

• Recognition while you speak. Nemotron now gets your voice as one continuous stream instead of phrase by phrase. It keeps the context of the whole utterance, so it recognizes noticeably better, and the text is ready about a quarter of a second after you stop, however long you talked.
• Live text above the recording pill shows what's been understood so far. Turn it off in Settings → Show text while I speak.
• Much faster on Intel 12th-gen and newer CPUs. Models use at most four threads; with eight, work spilled onto efficiency cores and Nemotron ran up to 18× slower.
• Nemotron 3.5 ASR · Max on the Models page: the same model without 8-bit compression, the most accurate option (2.6 GB).
• app.log in the data folder records what the app did — never what you said.

Windows: BeautifulVoice-Setup.exe · macOS (Apple Silicon): BeautifulVoice-macOS.dmg · portable: BeautifulVoice-windows.zip

Website: https://beautiful-voice.vercel.app

Beautiful Voice 0.2.0

Windows: download BeautifulVoice-Setup.exe and double-click it — it installs for your user in a few seconds, no administrator prompt. If SmartScreen warns about an unsigned app, choose More info → Run anyway. A portable BeautifulVoice-windows.zip is here too.

macOS (12+, Apple Silicon): download BeautifulVoice-macOS.dmg, drag Beautiful Voice to Applications. The app isn't notarized yet, so the first time right-click it and choose Open. Allow the microphone, and turn Beautiful Voice on in System Settings → Privacy & Security → Accessibility and Input Monitoring. The macOS build is new; please report anything that doesn't work.

What's new

• macOS build for Apple Silicon.
• In-app updates. When a new version is out, a card in the sidebar offers it. Installed Windows copies download the installer, update and restart by themselves; from now on you won't need to visit this page for updates.
• Settings → About → Check for updates.

Website: https://beautiful-voice.vercel.app

Beautiful Voice 0.1.1

Download BeautifulVoice-Setup.exe below and double-click it. It installs for your user in a few seconds (no administrator prompt), adds Start menu and desktop shortcuts, and can be removed from Settings → Apps. Prefer no installer? Take the portable BeautifulVoice-windows.zip.

The app isn't code-signed yet, so SmartScreen may ask you to confirm: More info → Run anyway.

What's new

• Windows installer with the Beautiful Voice icon on the app, its shortcuts and the setup itself.
• Models without Hugging Face. Every model is mirrored in this repository's models-v1 release. By default the app times both sources for a few seconds and downloads from the faster one; if either is unreachable, the other is used.
• Nemotron 3.5 ASR is the default model (32 languages, detects the language itself) and downloads by itself on first start.
• The interface speaks nine languages: English, 中文, हिन्दी, Español, Français, Português, Русский, Deutsch, 日本語.
• Half the size: unused Qt modules are gone — the installer is 89 MB.

Website: https://beautiful-voice.vercel.app

Beautiful Voice 0.1.0

The first release of Beautiful Voice — local dictation for Windows.

Download BeautifulVoice-windows.zip below, unzip it anywhere and run BeautifulVoice.exe. On first start the default speech model (Nemotron 3.5 ASR, ~650 MB) downloads by itself; after that everything works offline. The app isn't code-signed yet, so SmartScreen may ask you to confirm: More info → Run anyway.

What's inside

• Ctrl + Shift + Space in any app → speak → press again → the text is pasted where your cursor is. Esc or ✕ cancels.
• Phrases are recognized while you talk, so the text is ready about half a second after you stop.
• 11 local models: Nemotron 3.5 ASR (32 languages, default), Parakeet TDT v3/v2, Canary 1B v2, GigaAM v3, Nemotron Speech EN, Whisper Large v3 / Turbo / Small / Base, Distil-Whisper.
• Accuracy and speed on every model card, plus a voice test that measures both on your own voice and computer.
• History with a limit you choose, hold-to-talk or press-twice, tray, autostart, clipboard restore.
• Interface in English, 中文, हिन्दी, Español, Français, Português, Русский, Deutsch, 日本語; light and dark themes.

Website: https://qpsychocode.github.io/beautiful-voice/

Speech models

Mirror of the speech models Beautiful Voice downloads, so the app works without access to Hugging Face. The files are unchanged copies; manifest.json lists each file with its SHA-256 and the repository it came from.

These models are not part of Beautiful Voice and keep their own licenses:

Model	Authors	License	Source of the files
Nemotron 3.5 ASR Streaming 0.6B	NVIDIA	OpenMDW-1.1	csukuangfj2/sherpa-onnx-nemotron-3.5-asr-streaming-0.6b-1120ms-int8-2026-06-11 (original: nvidia/nemotron-3.5-asr-streaming-0.6b)
Nemotron Speech Streaming EN 0.6B	NVIDIA	NVIDIA Open Model License	csukuangfj/sherpa-onnx-nemotron-speech-streaming-en-0.6b-int8-2026-01-14 (original: nvidia/nemotron-speech-streaming-en-0.6b)
Parakeet TDT 0.6B v3 / v2	NVIDIA	CC-BY-4.0	istupakov/parakeet-tdt-0.6b-v3-onnx, istupakov/parakeet-tdt-0.6b-v2-onnx
Canary 1B v2	NVIDIA	CC-BY-4.0	istupakov/canary-1b-v2-onnx
GigaAM v3	Sber (salute-developers)	MIT	istupakov/gigaam-v3-onnx
Whisper Large v3, Large v3 Turbo, Small, Base	OpenAI	MIT	Systran and mobiuslabsgmbh CTranslate2 conversions
Distil-Whisper Large v3.5	Hugging Face	MIT	distil-whisper/distil-large-v3.5-ct2

ONNX exports by istupakov and the k2-fsa / sherpa-onnx project. Files over 2 GB are split into numbered parts; the app joins them and checks the SHA-256.