Luni·TTS v28 · conversational

100% in-browser conversational speech synthesis — no server. 9 emotions. ONNX Runtime Web + WebGPU.

Text

Set emotion inline with tags: [happy] … [sad] …. Add an intensity per tag with [emotion:NN] — e.g. [confused:40], [happy:80] (NN = 0–120%). A bare [emotion] uses that emotion's default. Each tagged span is generated separately and concatenated; the global slider scales them all.

Voice

ODE steps · 32

Seed

ODE io-binding on GPU (no per-step round-trip) — uncheck to compare

Initializing WebGPU…

Model: v28_rc1 (conversational, single-forward for clean speech). Emotion = style_emb added to the voice vector (style = voice + α·emotion); [emotion:NN] sets α per tag. confused/whisper default to 60% so they don't mumble. NV tags (<laugh>) land in v29.