A Method to Reduce Model-Base Volume for Speech Recognition
Jie Zhu · Audio Engineering · 2006
In the speaker independent isolated words recognition system, the limitation on the volume of model-base is a significant factor to hamper the improvement on recognition range and performance. The cost and computation time that a huge model-base requirs make it difficult for the recognition system to be used on a large scale. A method is presented to greatly reduce the of model-base volume in speech recognition applications for the speaker independent isolated words recognition system. In this method, the most suitable models is selected by using a process similar to different species fighting for food. Simulating results show that by use one general model instead several models in some cases, this algorithm can obviously reduce model-base capability requirement distinctively without significantly affect the recognition rate.