Apple quick start

The Swift package for iOS and macOS: the install line, where the key goes, and the quick start.

The Swift package runs on iOS and macOS, on Apple silicon. It uses the Neural Engine where the device has tested it against the exact CPU engine, and the CPU engine otherwise. The source is at github.com/Quran-Lab/tajweedformer-apple.

Install

// Package.swift, or in Xcode: File > Add Package Dependencies
.package(url: "https://github.com/Quran-Lab/tajweedformer-apple", from: "3.1.4")
// product TajweedFormer

import TajweedFormer

The key

An iOS app uses an apps key and its macOS build a desktop key: two keys. The app is identified by its bundle id. Keep the key out of your repository:

// Secrets.xcconfig (git-ignored), the target's base configuration
QURANLAB_KEY = qlk_v1_...

// Info.plist: QuranLabKey = $(QURANLAB_KEY)
let key = Bundle.main.object(forInfoDictionaryKey: "QuranLabKey") as? String

Add the App Attest capability: it proves your app's install, and the recordings learners share are not uploaded without it (recording sharing).

Client, model, listen, read

import TajweedFormer

func quickStart(key: String) async throws {
    // once per process
    let client = try TajweedClient(key: key)
    // downloads the model once, verifies it, opens and warms it
    let model = try await client.model()
    // asks for the microphone; 1.5 s of silence end the take
    let session = try await model.listen(options: .init(endOnSilenceMs: 1500))
    for await chunk in session.chunks {       // every 320 ms
        print(chunk.decoded.text)             // what the model heard so far
    }
}

Call it from a button: Task { try await quickStart(key: key) }.

  • The microphone. listen() asks for the permission, captures at the device's own rate and carries that rate to the session, so no sample rate appears here.
  • The end of a take. With endOnSilenceMs the take finishes by itself 1.5 s after the learner stops; without it, call try await session.finish().
  • The next take is a new model.listen().
  • Errors. Every error is a TajweedError with a stable code, a fix and a link into Errors.

What comes back

Each chunk carries chunk.decoded (the text, each token with its times and confidence, and how many tokens are settled) and chunk.heads (the raw outputs you selected). Build your own feedback says what each of them is.

The exact names for your SDK version are in the package's README.