It is highly recommended that a speech-recognition engine be scalable so that it can work with different processor speeds and memory configurations. This not only broadens the engine's market, but it provides experienced users with the ability to make tradeoffs between recognition accuracy and demands for processor resources.
When an engine is first installed or loaded, it should analyze the machine and make certain tradeoffs for optimal performance. To run effectively on slower machines, an engine should accept a higher error rate in recognitions.To take advantage of faster machines, the engine should:
· Use threads so the engine runs on symmetric multi-processor (SMP) machines that are running with Windows NT.
· Take advantage of a DSP if one is available.