SpeakEasy

Practice spoken English, one sentence at a time.

SpeakEasy is a SwiftUI app that uses Apple's new SpeechAnalyzer API (iOS 26) to score what the recogniser heard against French→English practice sentences in real time, with phonetic tolerance for common French-speaker pronunciation.

Features

• Listen — Hear the English sentence read aloud before you speak
• Record — Speak the sentence; progressive transcription streams as you talk
• Score — Per-attempt score (0–100) with token-by-token feedback (correct / missing / incorrect / extra)
• Progress — Tracks today's attempts, completed sentences, and difficult ones to review
• Settings — Adjustable session size

Requirements

• Xcode 17+ (Apple SpeechAnalyzer SDK ships with Xcode 26 toolchain)
• iOS 26.0+ device or simulator
• Microphone and Speech Recognition permissions

Setup

git clone git@github.com:S1933/SpeakEasy.git
cd SpeakEasy
open SpeakEasy.xcodeproj

Build and run on a real device for best results — the simulator has no real microphone and the new SpeechTranscriber assets may not be installed.

Permissions

Required Info.plist keys (already configured):

• NSMicrophoneUsageDescription — "SpeakEasy uses the microphone so you can practice spoken English."
• NSSpeechRecognitionUsageDescription — "SpeakEasy uses speech recognition to compare what you say with the practice sentence."

Architecture

SpeakEasy/
├── Features/
│   ├── Home/         Stats + Start CTA
│   ├── Practice/     Recording flow + per-sentence view
│   ├── Settings/     Session size
│   └── Summary/      End-of-session recap
├── Services/
│   ├── Speech/       SpeechRecognitionService, AudioSession, Permissions
│   ├── Scoring/      SentenceScoringService (token-level diff via Levenshtein-like DP)
│   └── Progress/     SwiftData persistence
├── Models/           LearningSentence, AttemptResult, etc.
├── Data/             sentences.json, SentenceRepository
└── Shared/
    └── Components/   MicrophoneButton, SentenceRow, etc.

Key technical decisions:

• SpeechAnalyzer pipeline — uses SpeechAnalyzer + SpeechTranscriber with the .progressiveTranscription preset for streaming results. Audio is captured via AVAudioEngine tap at the hardware format (typically 48 kHz Float32), then converted per-buffer to 16 kHz Int16 mono interleaved. A single reusable AVAudioConverter (with .noDataNow on dry input) is kept across callbacks — no allocation on the real-time audio thread.
• Concurrency — @Observable + @MainActor; SpeechRecognitionService is main-actor isolated; long-running work (analyzer.start(inputSequence:), result collection) is dispatched as unstructured Tasks with explicit error handling.
• Persistence — SwiftData with two models: SentenceProgress (per-sentence score history) and AppSettings (session size).

Tests

xcodebuild test -project SpeakEasy.xcodeproj -scheme SpeakEasy \
  -destination 'platform=iOS Simulator,name=iPhone 16'

24 unit tests cover scoring, normalization, and feedback.

License

Private project — all rights reserved.