iTantra On-Device STT

Translate into Speech model Decoder Threads

Best on desktop Chrome or Edge. Models download once (about 1 GB for two languages) and stay cached.

Upload a file

load models first

VADlanguage IDencoderdecodetranslate· bar = compute time per part

Compute on this device

0.0 s

speech processed

–

compute ÷ speech time

GMAC executed

requests sent while working

Stage	Time	GMAC

GMAC = billions of multiply-accumulates, counted from each model run's tensor shapes.