amivoice-realtime

Speech in, partial and final out

Unofficial. AmiVoice is a trademark of Advanced Media, Inc.

s command · sent once on connect

s MSB16K -a-general resultUpdatedInterval=1000 authorization=YOUR_ONE_TIME_TOKEN

Microphone

getUserMedia

resample()

16000 Hz

0samples out of 0

floatToInt16()

big-endian

MSB16Kmost significant byte first

p packet

0 sent

70 …

every 100 ms · 1600 samples

AmiVoice

not on this page

partialpartialfinal

Everything up to the last stage runs for real, using the package's own functions. Nothing leaves your browser.

The last stage is dark on purpose. Recognition needs a service ID and password, and a token endpoint on a public page would let anyone spend the account behind it. In your own app that stage is one createTokenCache call on the server.