TERRA-109READLive
Part numbers are where voice AI breaks.
Built so that mixed letters and digits come through correctly: brands and codes fed to the recogniser, a second listen to anything spelled out, and the recording used to correct the record afterwards.
"Two-three-five, sixty, three-five-three-hundred" has to become 235-60-35300 and nothing else. Even a speech model built for the job gets about 85% of such sequences right, so the desk does not trust one pass.
Inside TERRA-109 READ
5 capabilities, one number each.
TERRA-109-01Live
Your vocabulary, up front
Brands, product families, names and codes (up to fifty per call) are given to the recogniser before the call starts.
TERRA-109 READ
TERRA-109-02Live
Words it learns
A term heard the same way on two calls (once, for a code) joins the vocabulary, and anyone on your team can remove it.
TERRA-109 READ
TERRA-109-03Live
A second listen
The last minute of audio is kept. When a caller spells something or says a number, that stretch is transcribed again before the agent reads it back.
TERRA-109 READ
TERRA-109-04Live
The recording corrects the record
After the call, the recording is used to fix any misheard e-mail address or code on the call's record.
TERRA-109 READ
TERRA-109-05Live
Cyrillic look-alikes fixed
A К or А typed on a Russian keyboard inside a part number is turned into its Latin twin.
TERRA-109 READ