100% local speech dictation app with wakeword detection

Name: 100% local speech dictation app with wakeword detection
Availability: InStock
Author: computerex

by computerex·Mar 5, 2026·1 point·0 comments

Visit Project View on HN

AI Analysis

●●●BangerWizardrySolve My ProblemZero to One

Private wake-word dictation across three OSes outmaneuvering Whisper API, Google Docs Voice, and Otter.ai subscriptions entirely.

Strengths

•True local inference across Windows/macOS/Linux eliminates Whisper API costs and latency; rare cross-platform parity for speech models
•Wake-word activation (100+ phrases, custom options) solves hands-free UX gap in existing local solutions
•Custom Whisper+Parakeet inference in pure Go+CGO, optimized for CPU—genuine systems programming rigor backing the simplicity claim

Weaknesses

•$20 one-time pricing leaves scaling/maintenance path unclear; minimal public portfolio of accuracy vs. commercial baselines
•Gumroad distribution limits discoverability and limits feedback loops vs. open-source or App Store presence

Post Description

Hi HN! I built Evoke, a speech dictation app that runs 100% locally.

It supports both Parakeet and Whisper models, including Parakeet MLX on Apple Silicon, and works across Linux, macOS, and Windows (x86 and ARM).

One feature people seem to enjoy is wake word activation. You can trigger recording just by saying a phrase like:

“Dumbledore”

“Mr Anderson”

“Hey Jarvis”

Then just start speaking and your words are transcribed instantly.

Under the hood the app is written in pure Go with CGO, and the Whisper and Parakeet inference code is custom implemented and optimized for CPU performance, so it runs very fast even without a GPU.

This project was a real labor of love and a deep dive into speech inference and systems programming.