No emotion sliders. No language picker. No voice ID lookup. 30+ inline tags — emotion, nonverbal, prosody, language — written into your text the way a director writes stage directions. The 780M model reads them and performs.
[lang=en] Hey there — welcome back to The Local Frequency. [excited] We have got a really special show today, [laugh] I am still a little jet-lagged from the flight in. [pause=400ms] [lang=hi] तो चलिए शुरू करते हैं — [speed=0.95] सबसे पहले हमारे आज के मेहमान का परिचय। [lang=ja] [warm] 東京から、ようこそ。 [whisper] ……what a journey.
Four families. Inline. Compose anywhere. Same model reads them in any of 50+ languages.
"Hey there — welcome back. [excited] We have a special show today, [laugh] still jet-lagged. [pause=400ms] [lang=hi] तो चलिए शुरू करते हैं। [lang=ja] [warm] 東京から、ようこそ。 [whisper] ......what a journey."
emotion_id param. One emotion per call. No [whisper] mid-sentence.<break time="400ms"/> in some providers, undocumented in others.[laugh]: not supported. Generate, edit in Audacity.100K characters / month free. Apache 2.0 weights if you want to self-host. No card.