Feedback Assistant ID: FB24788469 (Cross-reference: Complementary to third-party TTS framework initiatives logged in FB23666342) Overview With the continued evolution of on-device machine learning and Apple Silicon's unified memory architecture, I propose that Apple evaluate expanding the native VoiceOver speech library in the upcoming iOS 28 cycle to incorporate lightweight, natural acoustic and neural voice models—specifically the profiles utilized across ChromeOS, Android, and CNET accessibility pipelines (including the natural Google Assistant voice profile). Hardware Feasibility & RAM Footprint • Demonstrated Efficiency on 8GB Systems: In real-world desktop testing across both an ASUS laptop and an HP laptop configured with 8GB of RAM, these voice profiles operate with near-instantaneous responsiveness, zero audio stutter, and negligible memory overhead. • Client-Side Execution Proof: In testing with NonVisual Desktop Access (NVDA)—an open-source Windows screen reader for blind and vision-impaired users—these voices run client-side using Google's WebAssembly (WASM) text-to-speech engine via background Chromium runtimes. Even during heavy UI tree navigation and multitasking, the 8GB systems run effortlessly without memory pressure warnings, UI latency, or process termination. • Viability Across 8GB & 12GB Apple Silicon: Apple Silicon devices leverage high-bandwidth unified memory and dedicated Neural Engine cores. Compiling or running these lightweight neural acoustic models via Core ML on all 8GB and 12GB devices running iOS 28—as well as future hardware generations—leaves substantial headroom for foreground apps without risking memory eviction. Technical & Practical Justification 1. Prosody and Auditory Fatigue: Modern neural synthesis captures natural cadence, warmth, and contextual phrasing, substantially reducing listening fatigue during multi-hour screen reading sessions compared to older formant or diphone synthesizers. 2. Deterministic On-Device Latency (Sub-50ms): Screen reader users require immediate auditory feedback when navigating character-by-character or word-by-word. Running these models locally on-device guarantees the sub-50ms response times essential for VoiceOver, with zero network latency and complete offline reliability. 3. Cross-Platform Continuity & Inter-Platform Synergy: Blind and low-vision users who switch between desktop environments (such as Windows with NVDA or ChromeOS) and iOS benefit greatly from vocal consistency across their workflows. Furthermore, given Apple's established collaboration with Google regarding foundational AI models, exploring speech asset licensing or compiling these lightweight voice packages into native iOS speech audio units represents a logical, accessible extension of existing synergy. Proposed Solution Provide downloadable, on-device voice packages for these high-efficiency neural and ChromeOS-style profiles directly within: Settings > Accessibility > VoiceOver > Speech > Voice in iOS 28.
It would be greatly appreciated if I could get community feedback. If posting in a language other than English, I'll use Gemini to translate it into English and post the translation for English speakers as a reply to this post. Also, if you want to check out my posts on Acapela's voice library integration, and my Expressive Siri posts, check out the links below and give me your thoughts on them as well. This post will be cross-linked on those as well, creating a closed triangular circuit. Expressive Siri Link: https://developer.apple.com/forums/thread/834920
Acapela Voice Library Integration Link: