I'm getting a very similar crash, with no obvious cause—just started today, as well, in a context that was previously stable. Super weird. Did you ever find out what was up?

Just to note (since I'm inexplicably no longer able to edit my question), I've since verified that other Bluetooth headphones demonstrate the same problem. So it seems to be specifically related to Bluetooth—presumably due to it being released when my app is suspended... Which does seem to be the purpose of CBCentralManager. So perhaps someone could just verify whether I should pursue finding a fix with Core Bluetooth.

I'm resurrecting this old question because I'm (finally) looping back around to this problem. I have a very rare crash during playback that I've been trying to debug, and noticed that I can't run with Address Sanitizer enabled because my app has this overflow when trying to create MusicTrackNewUserEvents (I do this at launch)—so it basically stops on that before I can even start playing anything. But what is the procedure for properly creating User Events that we can respond to as callbacks? I can imagine defining a "thin" struct for the 4-member [UInt8], but then how do I get the User Event's timestamp? In my original code, the User Event is wrapping our custom Event struct, which includes the timestamp and duration, but as I mention above, this requires Float64 (not UInt8). What we basically need is a timestamped User Event for signalling certain types of sequenced events during playback. We use that timestamp directly, so I either have to encode the time in the User Event or figure out how to get the time. If anybody has an example of how this is actually supposed to work it would be much appreciated.

Okay, I see I can do this much more simply. I've rewritten what I had based on Gene De Lisa's post here, on the AudioKit GitHub: https://github.com/AudioKit/AudioKit/issues/1393

Ditto. Total insanity... FIX THIS, Apple!!! Disabling network connection also seems better for me, so far.

Just to add an update to this: I was able to improve the "seam" in my seamless loop by manually setting the loop start to the priming frame count from afinfo, and the loop end to priming + valid. However, I'm now wondering whether a truly seamless loop is actually possible with compressed audio. Since the signal at any given point is always an approximation when encoded, the audio on both sides of a "cut" in the original wav must also be an approximation, meaning that the two sides won't necessarily be as "similar" as in the original. That seems reasonable to me, but is it correct? I'm talking about a "seamless" loop of an ambient soundscape recording, btw, so there's no actual silence (i.e., it's not a drum loop or anything like that).

Thanks for the response. Actually, it will pause and stay paused if I break in the main View, but in a service function I have for processing data it stops for a few seconds, then continues. It's a static function in a singleton, but I don't think that would make a difference. However, I actually just noticed a couple of things: 1) the app is actually detaching and continuing, and 2) it's posting an error that I hadn't noticed before (gulp): Message from debugger: The LLDB RPC server has crashed. You may need to manually terminate your process. The crash log is located in ~/Library/Logs/DiagnosticReports and has a prefix 'lldb-rpc-server'. Please file a bug and attach the most recent crash log.... It does the same breaking at various points in a custom package I'm developing alongside the app. I'll file a bug... I guess this should go to Swift, not necessarily Apple? EDIT: Actually, I see a similar unresolved issue from August at bugs.swift.org: https://bugs.swift.org/browse/SR-15110?jql=text%20~%20%22lldb%20crash%22 (In case anyone else is seeing the same issue.)

Thanks for the info @ConatusEng, but unfortunately I get a ModuleNotFound error when trying to import tensorflow_text. I've tried several different approaches, from various places on the internet, but no luck. I did notice that, for whatever reason, my conda install is miniconda3, not miniforge3, which a lot of people seem to have. I've running torchserve locally under a different env (for several months), and it works fine, so I don't really want to re-install conda if I can help it. Installing tensorflow-text from source does complete successfully, so I'm really not sure why the module won't load. Any thoughts?

I'm not sure that it's concerning, but it's annoying that we don't have anything yet. I have yet to try an LLM that's really good for Swift. Understandable. They're probably working to make sure they're substantively better than anything else out there.

Yeah, crazy... I could really use Swift Assist to deal with some SwiftUI problems I'm having. None of the LLMs are particularly good at SwiftUI, and seem to be wasting more of my time than they're saving. When, Apple???? You're the only ones with the data and the incentive to build a really knowledgeable Swift coding LLM.... super frustrating. This should be a no-brainer for Apple.

I hope WWDC gives us some news about Swift Assist. We really need a better LLM for Swift, imho.