

VoxMelt
Voxmelt - private voice-to-text for any Windows PC. CPU dictation via NVIDIA Parakeet v3 (no GPU required) plus an on-device AI text studio. Nothing is uploaded.
Cost / License
- Freemium (Subscription)
- Open Source
Platforms
- Windows
Features
Properties
- Privacy focused
Features
- Dictation
- Speech Recognition
- Local AI
- Voice to Text
- Offline
- Speech to text
VoxMelt News & Activities
Recent activities
- Oklonsica liked VoxMelt
- Elabosak added VoxMelt
Elabosak added VoxMelt as alternative to Wispr Flow, Superwhisper, Audiotype - Audio & Video Transcription and Vibe Transcribe
VoxMelt information
What is VoxMelt?
Private, local voice-to-text + on-device AI text studio for Windows.
Dictation runs on your CPU (NVIDIA Parakeet v3), AI polish on your NVIDIA GPU. Nothing is uploaded on cloud.
VoxMelt is private dictation for any Windows PC. Press a hotkey, speak, and NVIDIA's Parakeet v3 model transcribes you in real time on your CPU - no cloud, no per-minute fees, no GPU required. On VoxMelt's published real-clip benchmark the CPU engine wins on fast natural speech, the clip closest to real dictation (2.0% vs 3.1% word error), with no GPU, no VRAM and a 4-second load; Whisper large-v3 on a GPU is still ahead overall, which is why it ships in the app too. If you do have an NVIDIA GPU, a local LLM (via Ollama) rewrites, summarizes, or translates the text in one click across 17 templates and 82 tones, and a local TTS reads it back in 32 voice combos with WAV export. Audio and text never leave the machine: transcription, AI cleanup, and voiceover all run on hardware you own. Free tier (60 min/day) + 15-day full trial; Pro $9.99/mo, Studio $14.99/mo.






