ASR System Comparison on New High Performance Workstations
SVM HeaderParse 0.1
user correction - Legacy Corrections
In the last few years the computational power of general-purpose machines has dramatically increased; furthermore high performance workstations quite often offer multimedial capabilities. Therefore it is now possible to implement automatic Speech Recognizers (SRs) without dedicated hardware devices such as DSP boards. In this paper performance and costs for different configurations of some SRs based on workstations and mixed architectures (workstations equipped with DSP boards), developed at IRST, are presented and discussed. Although the response time depends both on the speech modalities, ( isolated words or continuous speech ), and the hardware platform on which the SR runs, nevertheless, for isolated words, considering the overall system price, a 486 machine can give very interesting results. 1 Introduction Integer and floating point performance of new machines based on the Hp PA-RISC and the DEC Alpha chips can be compared to mainframe computer capabilities. Furthermore, high per...