Neural Text-to-Speech

Checking…

Qwen3-TTS and Kokoro voices on MLX or CUDA backends. Streaming audio with Preston-Blair visemes.

Real-time factor—

Cache hit—

qwen3

ddx_adam

en · male · en-US

qwen3

ddx_bella

en · female · en-US

qwen3

ddx_heart

en · female · en-US

Engine

Language

Voice

Text

Emit visemes (lipsync timeline)Speed

1.00xViseme window

2.0s

Live waveform

Active visemesil

A/V sync

Audio @0.00s

Visemes @0.00s

Δ+0 ms