Voice Activity Detection 
and AGC
by D2 Technologies, Inc.

Software Overview

D2's Voice Activity Detection and AGC algorithm module detects voice activity and adjusts the voice 
signal to a target power level. The Voice Activity Detection algorithm adapts to background ambient 
or line noise as well as the presence of echo, classifying voice activity as "early" versus 
"sustained" and assigning an "effort level" to the speaker that is independent of network loss. The 
AGC algorithm adjusts the voice power level to a user specified target within a user specified time 
constant. The AGC is integrated with the Speech Detect algorithm to ensure that background noise or 
echo does not get amplified up to the specified target level.
Designed to the VP Open interface, the Voice Activity Detection and AGC module easily integrates 
with other VP Open modules, runs in real-time concurrently with other VP Open modules, and handles 
any number of voice ports assigned to the DSP. Like all of D2's algorithm modules, Voice Activity 
Detection and AGC consists of robust software that is rigorously tested

Features and Benefits

•	Voice Activity detector classifie voice activity as "early" or "sustained" according to user 
specified parameters
•	Adjusts voice signal to a specified target power level within a specified adaptation time
•	Voice Activity detector adapts to different background ambient noise or line noise, adapts to 
echo when used in conjunction with D2's network echo canceller
•	Accepts telephone-band codec or linear voice data
•	Assigns an "effort level" to the voice activity that is independent of the network loss or 
speaker level

Processor and System Specifications

•	Devices supported: TMS320C2xx, TMS320C3x, TMS320C5x
•	Processor resources required:
•	Algorithm category: Telecommunications, Audio
TMS320C2xx (Est.)
TMS320C3x (Est.)
TMS320C5x
Program memory
1 kwords (speech detect), 
0.5 kwords (AGC)
0.7 kwords (speech detect), 
0.3 kwords (AGC)
0.9 kwords (speech detect), 
0.4 kwords (AGC)
Data memory/
port
144 words (speech detect), 
8 words  (AGC)
120 words (speech detect), 
8 words  (AGC)
144 words (speech detect), 
8 words (AGC)
Processor loading
worst case
0.35 MIPS (speech detect), 
0.15 MIPS (AGC)
0.3 MIPS (speech detect), 
0.1 MIPS  (AGC)
0.3 MIPS (speech detect), 
0.125 MIPS (AGC)

Usage Limitations or Performance Considerations

•	The host, voice data input and output, memory allocation, and module to module communication 
interfaces must conform to VP Open as described in the VP Open Software Interface Specification.

Algorithm Verification

D2's Voice Activity Detection algorithm is validated through the following suite of tests:
•	Half-duplex speech detection tests with taped male and female speech at different noise levels
•	Full-duplex speech detection tests that simulate two way conversation tested with the echo 
cancellation active
•	VP Open interface validation
D2's AGC algorithm is validated through the following suite of tests:
•	Steady state tests over the dynamic range with different target levels
•	Attack and decay tests
•	Subjective voice quality test

Availability

Currently available for the TMS320C5x. A demonstration package is also available for the TI EVM50 
DSP board. Implementation for the TMS320C2xx and the TMS320C3x will be available 2Q 1996.

Company Background and Contact Information

Contact:	D2 Technologies, Inc..
		104 West Anapamu Street
		Suite J
		Santa Barbara,  CA 93101

		(805) 564-3424
		Fax:  (805) 966-2144
		e-mail:  sales@d2tech.com