SBAA490 Application note

SBAA490A December 2021 – April 2022 PCM6120-Q1 , TAA5212 , TAC5111 , TAC5112 , TAC5211 , TAC5212 , TLV320ADC5120 , TLV320ADC6120

2.2 VAD Parameters

Table 2-8 shows the parameters of the VAD algorithm. These parameters reside in the 32-bit wide coefficient memory (Book 0, Page 1, Page 7, page 8. and Page 9) of the device.

Table 2-8 List of VAD Parameters

VAD Parameter	Function, Description
Initial learning period (ILP)	This is the amount of time the VAD algorithm takes to adjust itself to the background noise environment, from the time instant VAD is turned on.
Hold over counter (HOC)	On detecting voice activity, the VAD algorithm generates an interrupt, if the interrupt is programmed to be active high, then the interrupt goes high (logic 1) on detecting voice and goes low (logic 0) when there is no voice. Before going low, the amount of time interrupt stays high after the voice activity ceases to exist is determined by the HOC count.
Wakeup wait (WW)	If VAD is programmed to be in Auto mode, on detecting voice, it will automatically turn on the ADC and start recording, simultaneously also checking for voice activity. Wakeup wait is the amount of time for which VAD is suspended after going to recording mode and thereafter resumed.
Threshold (TH)	Threshold controls the decision boundary of the nodes of the decision tree. A higher value will increase the node thresholds of all the nodes of the decision tree, thus reducing the likelihood of false positives. Similarly, a lower value for the threshold parameter decreases the node thresholds, which reduces the likelihood of false negatives.

Initial learning period: ILP is the amount of time the VAD algorithm takes to adjust itself to the background noise environment, from the time instant VAD is turned on. Equation 1 shows the computation of the VAD_ILP parameter.

Equation 1.

I n i t i a l l e a r n i n g p e r i o d (s) = \frac{{I L P}_{10}}{(256 \times 8000)}

where

ILP₁₀ is the ILP register value in decimal form

The default value (0x001F4000) corresponds to 1 s. Table 2-9 shows the registers that control the VAD_ILP parameter.

Table 2-9 Programmable Coefficient Registers for Initial Learning Period

Coefficient	Page	Register	Reset Value	Description
VAD_ILP	0x07	0x7C	0x00	ILP Byte[31:24]
	0x07	0x7D	0x1F	ILP Byte[23:16]
	0x07	0x7E	0x40	ILP Byte[15:8]
	0x07	0x7F	0x00	ILP Byte[7:0]

Hold over counter: On detecting voice activity, the VAD algorithm generates an interrupt, if the interrupt is programmed to be active high, then the interrupt goes high (logic 1) on detecting voice and goes low (logic 0) when there is no voice. Before going low the amount of time interrupt stays high after the voice activity ceases to exist is determined by the hold over counter count. Equation 2 shows the computation of the VAD_HOC parameter.

Equation 2.

H o l d o v e r c o u n t e r (s) = \frac{{H O C}_{10}}{(256 \times 8000)}

where

HOC₁₀ is the HOC register value in decimal form

The default value (0x00032000) corresponds to 100 ms. Table 2-10 shows the registers that control the VAD_HOC parameter.

Table 2-10 Programmable Coefficient Registers for Hold Over Counter

Coefficient	Page	Register	Reset Value	Description
VAD_HOC	0x08	0x0C	0x00	HOC Byte[31:24]
	0x08	0x0D	0x03	HOC Byte[23:16]
	0x08	0x0E	0x20	HOC Byte[15:8]
	0x08	0x0F	0x00	HOC Byte[7:0]

Wakeup wait: If VAD is programmed to be in auto mode, on detecting voice, it will automatically turn on the ADC and start recording, simultaneously also checking for voice activity. Wakeup wait is the amount of time for which VAD is suspended after going to recording mode and thereafter resumed. Equation 3 shows the computation of the VAD_WW parameter.

Equation 3.

W a k e u p w a i t (s) = \frac{{W W}_{10}}{(256 \times 8000)}

where

WW₁₀ is the Wakeup wait register value in decimal

The default value (0x01388000) corresponds to 10 s. Table 2-11 shows the registers that control the VAD_WW parameter.

Table 2-11 Programmable Coefficient Registers for Wakeup Wait

Coefficient	Page	Register	Reset Value	Description
VAD_WW	0x08	0x08	0x01	WW Byte[31:24]
	0x08	0x09	0x38	WW Byte[23:16]
	0x08	0x0A	0x80	WW Byte[15:8]
	0x08	0x0B	0x00	WW Byte[7:0]

Threshold: Threshold (TH) controls the decision boundary of the nodes of the decision tree. A higher value will increase the node thresholds of all the nodes of the decision tree, thus reducing the likelihood of false positives. Similarly, a lower value for the threshold parameter decreases the node thresholds, which reduces the likelihood of false negatives. Thus the threshold parameter can be adjusted to settle at the appropriate balance between false negatives and false positives.

Equation 4 shows the computation of the VAD_TH parameter.

Equation 4.

T h r e s h o l d_{n e w} = T h r e s h o l d_{d e f a u l t} \times 10^{\frac{t h r}{20}}

where

thr is the threshold value in dB (–20 dB to 0 dB )
Threshold_default is the default value in the Threshold register in decimal (16777216 )

The default value (16777216) corresponds to 0 dB. Table 2-12 shows the registers that control the VAD_TH parameter.

Table 2-12 Programmable Coefficient Registers for Threshold

Coefficient	Page	Register	Reset Value	Description
VAD_TH	0x09	0x50	0x01	TH Byte[31:24]
	0x09	0x51	0x00	TH Byte[23:16]
	0x09	0x52	0x00	TH Byte[15:8]
	0x09	0x53	0x00	TH Byte[7:0