A multi-modal virtual environment with text-independent real-time speaker identification

We present a speaker identification method to support a multimodal virtual environment. Most virtual environments require real-time and text-independent recognition of human speakers in order to manage the dialogue between the virtual characters and human users in the environment. Being widely used in pattern recognition tasks, neural networks have also been applied in speaker recognition. In this project, we developed a realtime text-independent speaker identification system based on probabilistic neural network (PNN). PNNs supply flexibility and straightforward design which make the system easily operable along with the successful classification results. We were able to correctly identify 96% of the speakers, using 0.8 seconds of test samples from each speaker. In addition to the description of the system and the experimental results, the effects of the feature vectors and codebook sizes on the performance are provided.