Hi everyone, With the initial rollout of the iOS 27 Beta, the advanced configuration sliders for the new, highly expressive Siri voice models are currently hidden/disabled on devices operating under an 8GB RAM threshold. While the local AFM 3 Core Advanced engine is memory-gated to 12GB+ hardware due to local footprint constraints, this creates a significant barrier for users who rely on next-generation vocal expressivity and fluid dictation features for accessibility. To address this, I have filed a formal feature request (FB23178983) asking Apple to route the expressive Siri framework through Private Cloud Compute (PCC) on 8GB devices (like the iPhone 15 Pro or base iPhone 16 models). By utilizing server-side rendering over secure, stateless cloud nodes, Apple could easily deliver these vital accessibility configurations without exhausting the local 8GB memory footprint. I’m curious if other accessibility testers and auditors on this board are looking for this exact routing path? If you or your users rely on advanced speech synthesis features and are currently locked out by the local RAM gate, please consider filing a duplicate request or adding your diagnostic logs to FB23178983 to help increase visibility with the Core Accessibility and Software Architecture triage teams. Looking forward to hearing your thoughts and experiences with the current voice-indexing behaviors on 8GB hardware!
Proposing Expressive Siri Offload via Private Cloud Compute (PCC) for 8GB Devices (Feedback ID: FB23178983)
If you have thoughts on native Acapela library integration, feel free to append them to the post linked below.
Link:
I'm in the United States, so please use English if possible.
I have a new post requesting Apple integrate low latency neural voices into iOS 28 and it deals with the voices found in ChromeOS and the GoogleTTS for NVDA plugin. I request community feedback on that post, FB24788469 and it would be greatly appreciated.
Link:
Title / Comment Header: Diagnostic Update: Dedicated 6.21GB Apple Intelligence Partition Verified on 8GB Hardware Hi everyone, Providing an update with exact system storage telemetry from an 8GB RAM device on the latest developer beta: • Exact Path: Navigate to Settings > General > iPhone Storage > View All. • System Breakdown: The device explicitly itemizes Apple Intelligence at 6.21GB, situated immediately before the base iOS system image (approx. 16GB). • Dedicated Neural Allocation: This distinct partition confirms that the operating system has reserved and staged 6.21GB specifically for local foundation models, semantic weights, and Apple Neural Engine execution graphs—not general temporary cache. • Architectural Reality: This proves that 8GB hardware comfortably accommodates the full on-device Apple Intelligence foundation layer alongside a complete 16GB OS installation without storage constraints or partition failures. • The Continued Barrier: Despite a dedicated 6.21GB machine learning payload running natively on the device, natural expressive speech synthesis remains locked behind a 12GB requirement in VoiceOver (com.apple.accessibility.heard). The local foundation infrastructure is already resident on flash storage. There is no technical justification to block VoiceOver from utilizing these models—either locally or by leveraging this resident framework to offload acoustic rendering to Private Cloud Compute (PCC) as detailed in FB23178983. Thank you to everyone tracking and sharing this thread (now past 2.7K views). Please continue referencing FB23178983 in duplicate feedback submissions to keep pressure on the software architecture and accessibility teams. Also, I have a new post, FB24,788,469 advocating for the inclusion of the same voices that the GoogleTTS for NVDA add on uses, the ChromeOS, which are the compact variants, and the CNet voices, which are the natural variants, which include the voices for Google Assistant and Talkback. I would greatly appreciate it if I got feedback on it as well. It already has 442+ views, and I just posted it yesterday.