piper.download_voices pastes the voice name straight into the HTTP request
line, and http.client encodes that as ASCII — so it dies with
UnicodeEncodeError on the ã in pt_PT-tugão-medium before a byte leaves the
container. Every pt_BR voice downloads fine. The failure lands precisely on
the one voice the pt-PT pair needs, and it is the *only* European voice in
the catalogue.
The entrypoint now falls back to fetching the model and its config itself
with the path percent-encoded, which is all the downloader was missing.
Same host, same files, same destination names, so the cached-voice check
and the server invocation are unchanged.
prosolis
3b714e297a