Beingpax/VoiceInk

VoiceInk – Offline, AI‑powered transcription for macOS

What it is – VoiceInk is a native macOS 15+ application that turns spoken words into written text on‑device, using local inference of OpenAI’s Whisper (via whisper.cpp) and other GGUF models. The app is positioned as a privacy‑first alternative to cloud‑based dictation services.

Key capabilities

• Instant transcription – Local AI models (e.g., SenseVoice Small) run on the Mac, delivering “almost instant” speech‑to‑text with the README’s claim of ~99 % accuracy.
• Full offline processing – No data leaves the computer; the app does not rely on external APIs.
• Context‑aware modes – The program can detect the foreground app or URL and automatically switch to a pre‑configured transcription profile.
• Custom dictionary & smart replacements – Users can teach the model new terminology and define text‑expansion rules.
• Global shortcuts – Keyboard or mouse shortcuts (including push‑to‑talk) are configurable via the KeyboardShortcuts library.
• Built‑in AI assistant – A “ChatGPT‑like” conversational mode is bundled, powered by the same local models.
• Smart modes for writing styles – Quick toggles between transcription presets optimised for different contexts (e.g., code, prose, emails).

How it works under the hood

• Speech engine – whisper.cpp provides fast, CPU‑only inference of the Whisper model.
• Model handling – Transcribe-cpp-swift (a SwiftPM wrapper around transcribe.cpp) loads GGUF‑format models such as SenseVoice Small.
• Audio capture – FluidAudio supplies low‑latency audio input.
• App integration – Libraries like SelectedTextKit read the current selection, while MediaRemoteAdapter pauses media playback during recording.
• Updates – Sparkle handles automatic app updates for the binary distribution.

Getting started

• Download – A free trial (with optional paid license for updates & support) is available at https://tryvoiceink.com.
• Homebrew – brew install --cask voiceink installs the pre‑built app.
• Build from source – Follow the BUILDING.md guide; the repository includes a Swift‑based macOS project that links the above native libraries.

Community & contribution

• The repo is open‑source under GPL‑v3. Pull requests are currently disabled, but users may fork the project, file issues, suggest features, or improve documentation.
• Support is offered via a Discord channel and email for license holders.

Why it matters VoiceInk demonstrates that high‑quality speech‑to‑text can be delivered entirely offline on a consumer laptop, addressing privacy concerns that dominate many cloud dictation services. By bundling Whisper‑style models with a polished macOS UI, it provides a usable reference implementation for developers interested in on‑device AI transcription.

---

TL;DR – VoiceInk is an open‑source macOS app that transcribes speech to text locally using Whisper‑derived models, offering privacy‑first, instant dictation with context‑aware presets, custom dictionaries, and a built‑in conversational assistant. It can be installed via Homebrew or built from source, and is licensed under GPL‑v3.