Text normalization for the formant TTS (sound.speech).
Turns a raw string into a flat token list the G2P layer consumes:
{ word = "hello" [, emph = true ] } a lowercase word (a-z + apostrophe)
{ punct = "." | "," | "?" | "!" } a phrase/sentence boundaryHandles: number expansion (integers, decimals, negatives — "42" -> "forty
two"), a small abbreviation table, *word* emphasis markup, and punctuation
classing (.;:!? and em-dashes become boundaries; everything else is dropped).
| Name | Signature |
|---|---|
| number_words | number_words(s) -> words |
| normalize | normalize(s) -> tokens |
number_words(
s) ->words
Expand a number literal into English words
s is a decimal literal ("42", "3.5", "-7"). Returns an array of
lowercase words ({ "forty", "two" }). Integers are expanded up to the
billions; fractional digits are spelled out one by one after "point".
normalize(
s) ->tokens
Normalize a text string into word/punct tokens