Readaloud skips chapters (audio missing) when transcribing with large-v3 on a remote whisper.cpp server

I import an ebook and audiobook and all chapters show up correctly, but after creating the readaloud some chapters have no audio at all - I can read them but readaloud doesn't work on them. I'm transcribing with large-v3 against a whisper.cpp HTTP server I run in Docker Desktop on a more powerful machine and point Storyteller at over my LAN. The same book aligned locally with tiny had all its audio. Storyteller 2.14.14.

Resolution: Two things combine here. First, non-turbo large-v3 is noticeably more prone to repeating spirals during transcription, and a spiral usually causes a whole chapter to be dropped from the readaloud. Second, an arbitrary upstream whisper.cpp container is not necessarily the same build Storyteller expects, which can change results. Switch to large-v3-turbo - it is both faster and much less prone to this, and it resolved the missing chapters immediately here. For remote transcription, run the ghost-story wrapper rather than a bare whisper.cpp image: it pins the exact whisper.cpp version Storyteller uses, and with it even large-v3 produced a complete readaloud. Note ghost-story has no Docker container and does not run natively on Windows - run it under WSL, or follow the compiled-whisper-server community guide for a native Windows build.