Two voice tiers, both running on silicon we own. One of them can laugh.
Tier one
HawkTalk Gooder
54 voices, swappable per sentence
Faster than realtime — 1.45×
First take, every take
cannot do non-verbals
$0.02 / min ElevenLabs $0.17
Tier two
HawkTalk Amazing
Gasps, sighs, laughs, breaths
Yawns, groans, chuckles
3B model, expressive prosody
For the lines that carry weight
$0.04 / min ElevenLabs $0.17
The difference, in three seconds
Every clip below is the same sentence in both tiers. Gooder reads it.
Amazing performs it — the <tags> are model
vocabulary, so it produces the actual gasp, the actual sigh.