Breadcrumb navigation

NEC Voice Recognition

* This technology is still under research.
* Currently deployed only in Japan

Japanese

This technology extracts characteristics inherent in speakers from their voices to identify them.
Every person’s vocal organs such as his or her mouth, throat, and nose are like musical instruments. Because of differences in their shapes and movements among individuals, unique characteristics arise.

Features of voice recognition

Easy to deploy

Voice recognition can be performed on smart phones or tablets using a general-purpose microphone.

Easy to use

Speakers can be identified regardless of utterance content and language.
Recognition over the telephone is possible.

Recognition method

Voice recognition is carried out either based on recorded voice data for set phrases or on recorded natural, unstructured voice data.

Text-dependent recognition

In this method, utterance voice data for set phrases are recorded and used for recognition.
Matching is possible for short (0.5-3 sec.) voice data.
A different set phrase can be used for each speaker.
Since the voice data library is relatively small, this algorithm can be embedded in devices.

Text-independent recognition

In this method, utterance voice data for natural speech are recorded and used for recognition.
This requires a relatively long (10 sec. or longer) voice data.
Recognition is possible regardless of speaking speed, accent, or language.