In practice, for the best recognition performance, the recognition program has to be optimized for environment and configurations. For example, the background noise levels are different in different application scenarios. Different microphones have varying voice capture performance, which may lead to various voice levels.
To cope with these adverse factors, the parameters of the recognition program have to be adjusted for performance optimization. For most application-oriented projects, we need to run the cognition program at the target hardware platform and verify its performance in realistic scenarios at different parameter values. The traditional design-implementation-validation loop is time-consuming, as it may involve a huge amount of coding and recoding. Automatic code generation and adjusting parameters on the fly are the promising features of the BeagleBoard rapid prototyping, which not only allow us to reduce the development time, but...