Processing fails with "has no readaloud filepath after processing" when a quantized whisper model (tiny.q5) is selected

On server 2.14 I tried aligning an epub + m4b pair with several whisper models. With tiny.q5 processing failed outright with an error about the readaloud having no path after processing. The epub and audiobook both work fine on their own.

Resolution: Selecting a quantized whisper model such as tiny.q5 hit a bug in the model-selection logic, so the transcription step never produced usable output and processing aborted with the generic "has no readaloud filepath after processing" error. Switch the whisper model to a non-quantized variant (plain tiny, tiny.en, or a larger model) until you are on a build containing the fix. Be aware this error string is a catch-all for "processing ended without a readaloud" and has several unrelated causes — notably a transcoded-audio directory containing more than one codec — so if a non-quantized model does not clear it, clear the book's cache files and check the earlier lines in the log for the real failure. Reports of this error rose sharply after 2.14, and multiple distinct bugs behind it were fixed separately.