NEC Enables Voice Control of Smartphones in Noisy Environments

May 9, 2013
Mami Akasaka, Tech-On!

NEC Corp developed a speech recognition technology that enables to remotely control a smartphone, tablet computer, etc with voice even in a noisy environment.

The technology was developed by combining the company's noise cancellation and speech enhancement technologies.

While voice control functions have been adopted for home appliances and mobile devices, it is currently necessary to clearly speak to an embedded microphone at close range. Therefore, in noisy environments such as rooms with a TV, kitchens and vehicle interiors, mobile devices often do not respond to voice and respond to noise.

With the new technology, a mobile device accurately detects voice pronounced at a distance even in a noisy environment. For example, it becomes possible to control a device on a table from a distance of about 1m even in a noisy living room.

The details of the noise cancellation and speech enhancement technologies are as follows.

Noise components detected by sub-microphone to accurately follow noise

Many mobile devices including smartphones have a main microphone on the front side and a sub-microphone on the back (or lateral) side. This time, NEC made use of this two-microphone system.

With a two-microphone system, the main microphone and sub-microphone tend to mainly receive voice input and noise, respectively. And the new technology clarifies the components of the noise collected by the sub-microphone when the main microphone is receiving no voice input (such as a moment right before the beginning of voice input).

As a result, it becomes easier to accurately follow and remove noise even when the main microphone is receiving voice input.

Voice models used to modify sound

NEC developed a technology to emphasize the components of voice that is difficult to be recognized due to distortion caused by noise removal by using voice models. It modifies distorted voice into a voice model having a similar waveform so that the voice can be recognized. It realizes a speech recognition accuracy higher than that of ETSI Advanced Front-End (AFE), a global standard of speech recognition for mobile devices, the company said.