← Back

Promptur

Built

Signature capability

The lazy version of this, a loudness threshold plus last-word matching, is what the free clones ship, and it breaks in any room with a fan. The voice gate and the aligner both needed real signal processing to work.

What it is

The script floats in a strip just under the camera so you read almost straight down the lens. On-device speech recognition works out where you are in the text and scrolls to match, rather than at a fixed rate. Built with zero third-party dependencies and no network code at all.

Highlights

  • 12,333 lines of Swift across 44 files with no third-party dependencies.
  • 223 tests across 13 suites covering the aligner, scroll model, voice gate, reading pace and recognition delay.
  • An offline harness replays recorded audio through the real recogniser to measure tracking accuracy, rather than mocking it.
  • No account, no server, no telemetry and no network code, so the script never leaves the machine.

Tech · Swift, SwiftUI, AppKit, on-device SFSpeechRecognizer, AVFoundation, SwiftPM, XCTest.