User Manual

Unified Communications Microphone & Speaker System
YVC-1000 White Paper
17
1. Human Voice Activity Detection (HVAD)
1.1 Outline and purpose
HVAD is a technology which discerns whether or not human voices are included in voice signals that are picked up
by YVC-1000.
1.2 Operations and mechanisms
HVAD operates in the basic signal processing technologies in the three following functions to dramatically enhance
accuracy.
● Automatic tracking
● Noise reduction
● Automatic gain control
This section describes the HVAD mechanisms in combination with the above signal processing technologies using
the signal processing flow charts.
* For details about individual signal processing technologies, refer to the corresponding pages in Chapter 2.
1) In combination with automatic tracking
YVC-1000 can pick up voices clearly even in an environment where voice locations change among participants. An
audio source location is estimated by the microphones array control device that consists of three microphone
elements. When the sound source orientations are presumed, HVAD determines whether those sound sources are
human voices or not and uses those results to capture the orientations of the isolated sounds and steady noises as
the speakers’ locations to dramatically reduce mistaken identifications.
2) In combination with noise reduction
The noise reduction function is required to estimate noise components in order to eliminate only the noise
components from voices. The method to estimate a steady signal as a noise component is commonly used, but
some audio signals (such as prolonged voice "Aaa..." in conversation and music) may be eliminated as misidentified
noise components. Since YVC-1000's noise reduction can distinguish human voice components from noise
components using the HVAD determination results, noise reduction performance is improved.
3) In combination with automatic gain control
The automatic gain control function properly corrects the sound level of voices picked up by a microphone. This
requires the function to accurately pick up the volume of human voices before they are corrected. YVC-1000 can
estimate the level of human voices via HVAD with high precision by distinguishing between signals containing
human voice and signals consisting of only noises. The result is that auto gain control performance can be
Chapter 3
Two Unique Technologies to Enhance Six Sound Processing Capabilities