# privacy-first voice-to-text desktop

OiPerDESKTOP v2.0

Hold a global hotkey. Speak. Release.
Whisper processes audio locally — no cloud, no lag, no data leaks.
Text injects directly into any active application.

1.5s

avg transcription

0KB

cloud data sent

2.18×

faster than APIs

oiper — bash — 80×24

$ ls ./core-features/

Core Modules

hotkey.init()

Global Hotkey

Press-and-hold activation across every app, every window. No context switch.

privacy.local()

Local Inference

Whisper runs on your hardware. Audio never leaves the machine unless you choose.

backend.gpu()

GPU Acceleration

CUDA and Metal supported. Auto-detect or force-select your backend.

llm.cleanup()

LLM Text Cleanup

Gemini Flash Lite, local models, any OpenAI-compatible API for text polish.

inject.active()

Direct Injection

Transcribed text lands in your focused input. No clipboard. No paste.

model.manage()

Model Manager

Choose Whisper model size. Download, switch, and delete from built-in UI.

$ benchmark --audio=30s --lang=en

Benchmark

Tested on 30 seconds of English audio. Native code wins — every time.

> Native binary — no Python overhead

> GPU-accelerated Whisper inference

> Optimized local audio pipeline

> Zero network round-trips

★ OiPer Desktop1.5s

Lemonfox API3.27s

Python Faster-Whisper3.55s

OpenAI Whisper 1 API6.46s

$ cat privacy.conf

Privacy Architecture

[SECURE]

On-Device Audio

Recording is captured and processed locally. Never persisted by default.

[SECURE]

Local Transcription

Whisper runs on your CPU or GPU. No API calls, no data egress, ever.

[OPT-IN]

Cloud LLM Cleanup

Text polishing via cloud is entirely optional. Requires your own API key.

> SYSTEM READY

Own Your Voice.
No cloud required.

Fast, local, and entirely under your control.