Resource Requirements

Vestec ASR utilizes an efficient, scalable client-server architecture that can support thousands of ports given appropriate computational resources. (One “port” refers to one communication “channel”. So, a 4-port setup means simultaneous speech recognition on 4 channels.)

The following table presents sample memory and CPU usage for different number of speech ports (ie. channels) for Tier-1 ASR. The test was conducted on a machine with AMD Athlon II X2 245 dual-core processor, 4GB of RAM, and Linux CentOS 5. The grammar consisted of 500 distinct keywords while the processed audio was 1.6 seconds long. Each test ran for 140 seconds, with each interaction taking an average of 14 seconds.

NUMBER OF PORTS	MEMORY USAGE (MB)	CPU USAGE (%)
0	291	0
1	296	1
2	300	2
4	305	5
8	315	10
16	334	27
24	362	35
32	390	49
48	438	76
64	489	90