sibilant | detect speaking events from webRTC audio | Runtime Evironment library

I think you're mis-reading Table 2 from the article:

'#' One or more vowels

'*' (star) One or more consonants

'*' (circle) One of B, D, V, G, J, L, M, N, R, W, and Z: a voiced consonant

'$' One consonant followed by an E or I

'%' One of (ER, E, ES, ED, ING, ELY): a suffix

'&' One of (S, C, G, Z, X, J, CH, SH): a sibilant

'@' One of (T, S, R, D, L, Z, N, J, TH, CH, SH): a consonant influencing the sound of a following long u (cf. rule and mule)

'^' One consonant

'+' One of (E, I, Y): a front vowel

':' Zero or more consonants

The caption for that table says:

Special Symbols Appearing in the English-to-IPA Translation Rules

In other words, those aren't rules, but rather descriptions of the notation.

The real rules start on page 51 of the article. As I understand the algorithm, it will start by parsing the 't', and go to the TRULE.ENG section (page 58). The only rule that applies there is [T]=/T/. Then it parses the 'a', and looks in ARULE.ENG (page 51). Scanning the rules there, the first one I see that applies is [AU]=/AO/, which gives you the "au" as in "taught". From there it goes to the GRULE.ENG list and produces / / for "gh", and then produces /T/ again for the final "t".

As for "laugh", from the rules I'd say that it's going to pronounce the "au" as in "taught". And because the rule is #[GH]=/ /, there will be no other sound. "laugh" will probably sound like "law".