thermal yap · Windows 10/11 · CPU dictation

Built for Windows 10 and 11. Melt local.

Dictation runs on your CPU with NVIDIA's Parakeet v3 model, and voiceover runs on your CPU too: no graphics card required. Got an NVIDIA card (GeForce RTX, workstation RTX A-series / Ada, or a data-center A100 / H100)? It unlocks the local AI text studio while your voice stays on the processor. No NVIDIA card just means the AI studio waits - dictation works everywhere. Macs sit this one out.

$ nvidia-smi --query-gpu=name,memory.total --format=csv
NVIDIA GeForce RTX 3060, 12288 MiB
Runs on any Windows 10/11 PCdictation + voiceover on CPU · AI studio: NVIDIA
Get it from Microsoft Store

Direct installer, for PCs without the Store; Windows will warn because it is not code-signed.

Windows 10/11 · ~12 MB on disk · audio and text stay on your PC · what the app sends
install steps

What happens after you click

Microsoft Store recommended

No Windows warning: Microsoft signs the Store package. Free to install. Click Install, then Open: the app does not start by itself.

Direct download (5 MB, not code-signed)

Windows will warn. The way past each one:

  • If Edge says the file "isn't commonly downloaded", open the ... menu, then Keep, Show more, Keep anyway.
  • If "Windows protected your PC" appears, click More info, then Run anyway.
  • If Windows blocks it with no Run anyway (Smart App Control), use the Store instead.

Before your first dictation, either door

Voxmelt downloads its speech engine and model, about 840 MB: about 2 minutes at 50 Mbps, about 11 at 10 Mbps. Keep about 1.0 GB free while it sets up. About 1.0 GB stays.

Stuck? Write to help@voxmelt.com.

thermal budgeting · optional AI studio

Box copy lies. Do the VRAM math.

Planning the NVIDIA-powered AI text studio? Match your card against a cleanup model and see the fit check: both stay loaded, take turns, or pick something smaller. Hardware honesty over marketing vapor. Dictation needs none of this - it runs on your CPU.

compatibility check

Is your GPU ready to cook?

NVIDIA only · Win 10/11
Whisper 3 GB
4B 2.6 GB
0 GB6 GB12 GB
BOTH MODELS COEXIST · OVERCLOCKED W
Whisper stays warm. Zero handoff latency between recording and AI cleanup. Tensor-toasted throughput.
12 GB total VRAM3 GB Whisper2.6 GB Gemma 3 · 4B6.4 GB headroom
System requirements

What your rig needs

ComponentMinimumRecommended
OSWindows 10 64-bit (build 19041+)Windows 11
CPUAny modern 4-core x64 CPU (runs dictation + voiceover)8-core for extra headroom
RAM8 GB16 GB
MicrophoneAny input deviceUSB or headset mic
GPU · optional, AI text studio onlyNone for dictation & voiceover · NVIDIA 6 GB+ VRAM (Ampere or newer) for AIRTX 30/40/50-series · 8-12+ GB VRAM
NVIDIA driver525.0+ (GPU features only)570.0+ (req. for Blackwell)
LLM runtimeOllama on 127.0.0.1:11434 (AI studio only)Ollama, latest
Disk~12 MB app + about 1.0 GB engine & model on disk (an 840 MB download) · about 1.0 GB free during first setupSSD
NetworkAbout 840 MB before your first dictationThen licence checks and updates
Release archive