BetaEchophrase is a privacy-first, GPU-accelerated speech-to-text dictation app that runs locally on your native OS with versatile options.
Private. Fast. A polished, lightweight experience - unlimited & 100% offline after a one-time model download, no account needed. Native Tauri+Rust, not bloated Electron.
Free forever for personal use — no paywall. Commercial use requires Pro.
No cloud uploads - Self-serve GDPR friendly.
Read our Privacy Policy.
Why Echophrase?
Professional AI voice-to-text transcription without compromising your privacy
Your Voice Stays Yours
Transcription runs 100% on your device, unlimited, no account needed. Works fully offline after a one-time model download. Fully GDPR compliant - your audio is never uploaded, stored, or trained on.
Lightweight & Fast
GPU-powered transcription in under a second with recommended models. Native Rust + Tauri on Windows and macOS - a 34 MB install that idles lighter than a browser tab, not an Electron app.
Paste at Cursor
Speak anywhere text goes: hotkey, transcribe, and it lands at your cursor. Perfect for AI chats and code.
Cleans Up After You Pro Pro feature
An optional post-transcription local transform model cleans up grammar and punctuation after you speak - still fully on-device.
Your Format, Your Rules Pro Pro feature
Customizable JSON output with custom metadata for power users - your transcript, structured your way. Bring your own model: coming soon.
Made for Everyone
Full GUI localization - English and Traditional Chinese to start - plus UI scaling accessibility settings.
Email us with accessibility requestsSee it in action
Global hotkey support
Automatic clipboard copy
Choice of multiple ONNX / safetensors models
Push-to-talk & open-mic modes
One-click in-app model downloads
Featherweight by design
Built on Tauri + Rust, not Electron - native on Windows and macOS (Linux in alpha), and it shows in every number we measured
34.2 MB
Tiny install
The entire Windows app is ~6x smaller than the ~200MB runtime a typical Electron app ships before any app code.
~226 MB
Light idle RAM
Total RAM with the AI model loaded and ready to transcribe - always ready without weighing down your system.
+7.8 W
Barely a blip
A single transcription peaks at just +7.8W GPU and +4.3MB RAM, then settles right back to idle.
Flat
No memory creep
Back-to-back transcriptions hold flat after a one-time warm-up - a footprint that lasts all day, not one that degrades.
Measured with NVIDIA Parakeet-TDT 0.6B v3 loaded. See full benchmarks & methodology
How Echophrase Compares
Honest comparisons with other dictation and voice typing apps — including the free ones.
Ready to get started?
Download Echophrase today and experience private, GPU-accelerated AI voice-to-text
Windows 10/11 • DirectX 12 GPU
Mac (Apple Silicon) in beta • Linux GPU acceleration for Linux will arrive later in open alpha
No GPU (or Linux)? CPU still works surprisingly well.