Read-aloud generation fails downloading the whisper.cpp "legacy" build on an old Xeon CPU

Every attempt to generate a read-aloud fails during the transcription step. The server logs show it failing while fetching the whisper.cpp binary. I'm running the Docker image on a home server with an older Intel Xeon CPU (L5640) and the transcription engine set to OpenBLAS. Switching the engine to plain CPU gives the same error.

Resolution: Storyteller detects CPUs that lack the newer SIMD instructions whisper.cpp uses by default and downloads a "legacy" whisper.cpp build instead. A bug in that code path appended -legacy twice to the filename, so the URL pointed at a file that does not exist and the download always failed. That filename bug is fixed in 2.14.18, so upgrade to at least that version. Two additional things to check: don't use the OpenBLAS variant (it is a separate image, it is not meaningfully faster, and it has its own download failure on some CPUs) and if you're running under a hypervisor such as Proxmox, make sure the guest is not masking CPU features by emulating a lower-spec CPU model. On genuinely ancient hardware (Westmere-era Xeons and similar) transcription will be impractically slow even once the download works; the supported answer there is to offload transcription to another machine with ghost-story, following the "offloading transcription" tutorial in the Storyteller docs.