Pronunciation

The lexicon maps written terms to spoken forms so the narrator says jargon right. Matching is case-sensitive on whole words.

Packaged default

The default ships in the package (sase_listen/data/lexicon.yml):

Term	Spoken as
TUI	T-U-I
xprompt	ex-prompt
xprompts	ex-prompts
chezmoi	shay-mwah
uv	you-vee
PyPI	pie-pee-eye
mypy	my-pie
LUFS	loofs

SASE is deliberately absent until its pronunciation is decided.

Your own entries

Point lexicon: in the config at a YAML file of term: spoken form lines:

Kubernetes: kube
Grafana: gra-FAH-na

Your file merges over the packaged default, so it can add terms or override them. The merged sha256 is recorded in each episode manifest, so a lexicon change re-renders affected chunks instead of reusing stale cache audio.

The lexicon applies to spoken text only, never to ID3 or feed chapter titles. When a name still sounds wrong, fix the script wording first (write symbols as words, expand the acronym), and reach for the lexicon second.