Epic: Read Aloud (TTS) — speak selected text via hotkey (local + cloud)

[thefourcraft]

Read Aloud — the mirror of dictation

Press a hotkey -> grab the currently selected text -> speak it aloud, using a local neural voice or a cloud provider. Parallel subsystem mirroring the existing STT stack (CloudProvider/registry, APIKeyManager/Keychain, HotkeyManager, SelectedTextService, SoundManager).

Decisions (locked)

• Local engine: sherpa-onnx + Kokoro-82M (Apache-2.0, whisper.cpp-style C lib, neural, offline).
• Cloud providers: Deepgram Aura-2 (default), Inworld TTS-1.5 Mini, ElevenLabs v3, Gemini 3.1 Flash TTS, OpenAI gpt-4o-mini-tts, Cartesia Sonic-3.5. (All verified current as of 2026-06.)
• Playback: one shared progressive-PCM pipeline (AVAudioEngine) + whole-file AVAudioPlayer fallback.
• Text capture: existing SelectedTextService.fetchSelectedText() (AX read -> synthetic-copy fallback). One Accessibility grant already covers it.

Work breakdown

Foundation

• TTS: core provider abstraction (TTSProvider protocol + registry + routing) #207 Core provider abstraction (TTSProvider protocol + registry + routing)
• TTS: streaming audio output pipeline (AVAudioEngine PCM + AVAudioPlayer fallback) #208 Streaming audio output pipeline (AVAudioEngine PCM + fallback)
• TTS: trigger flow — hotkey + selected-text capture + speaking state #209 Trigger flow (hotkey + selected-text + speaking state)
• TTS: local engine — sherpa-onnx integration + Kokoro-82M model manager #210 Local engine: sherpa-onnx + Kokoro-82M model manager

Cloud providers

• TTS provider: Deepgram Aura-2 (default) #211 Deepgram Aura-2 (default)
• TTS provider: Inworld TTS-1.5 Mini #212 Inworld TTS-1.5 Mini
• TTS provider: ElevenLabs v3 (premium) #213 ElevenLabs v3 (premium)
• TTS provider: Google Gemini 3.1 Flash TTS #214 Google Gemini 3.1 Flash TTS
• TTS provider: OpenAI gpt-4o-mini-tts #215 OpenAI gpt-4o-mini-tts
• TTS provider: Cartesia Sonic-3.5 #216 Cartesia Sonic-3.5

UI + release

• TTS: settings UI — provider/voice/speed picker + API key entry #217 Settings UI (provider/voice/speed + key entry)
• TTS: speaking overlay + permissions/onboarding polish #218 Speaking overlay + permissions/onboarding
• TTS: release v2.0 — version bump, changelog, build + notarize #219 Release v2.0 (version bump, changelog, notarize)

Notes / corrections

• Chatterbox deferred to a later phase: it is 0.5B (not 350M), ~1–3 GB (not 700 MB), has no native Swift binding (sidecar/ONNX only), and watermarks output. Kokoro is the right first neural local engine.
• Release gating: notary issuer ID + Sparkle EdDSA key are open blockers per release-readiness notes.