Figure 6.20-4 Vad Decision Tree - Nuvoton ISD94124BYI Technical Reference Manual

STAT
NO
Note:
STAT = VAD_CTL0[7:0]
LTAT = VAD_CTL0[19:16]
STP = VAD_STATUS0[15:0]
LTP = VAD_STATUS1[31:16]
STTHREHWM = VAD_CTL1[15:0]
LTTHRE = VAD_CTL2[31:16]
DEV = VAD_STATUS1[15:0]
DEVTHRE = VAD_CTL3[15:0]
ACTIVE = VAD_STATUS0[31]
6.20.5.7 Tuning VAD Performance
This section provides a tuning guide for VAD performance. The VAD calculates power of an input
signal to determine whether any voice is present or not. The calculations include a combination of
long-term / short-term power and deviation power.
Step 1: Set Short Term and Long Term Power Attack Time
Short term power attack time should be properly set such that the short-term power reflects the
human voices. Long term power attack time should be set such that the long-term power represents
the ambient condition. In general, lower setting means slower attack or long responding time;
whereas higher setting means faster attack or short responding time. Suggested default attack time
setting:
Short term power attack time: STAT (VAD_CTL0[7:0]) = 0xAA. If slower attack is desired, e.g., if
the VAD is triggered frequently by many non-voice sudden sounds, 0x99 may be used. If faster
attack is desired, e.g., if the VAD appears to suffer long latency when clear voices are present,
Sep 9, 2019
ISD94100 Series Technical Reference Manual
ACTIVE=0
Short Term Power
Formula
To get STP
STP >STTHREHWM
YES
LTP
STP
DEV
=|LTP-STP|
DEV >DEVTHRE
YES
ACTIVE=1

Figure 6.20-4 VAD Decision Tree

Page 882 of 928
Long Term Power
Formula
To get LTP
NO
LTP> LTTHRE
YES
DEV
=|LTP(LTTHRE)-STP|
NO
LTAT
LTP
(LTTHRE)
Rev1.09