Voice-Based Recognition System for Non-Semantics Information by Language and Gender

The human voice not only provides information about the semantics of spoken words, but also contains voice information based on its characteristics. This paper designed feasible identification system for non-semantics voice information by language and gender, which are the two most important in voice signals. The proposed system is speaker-independent and text-independent: it fuses the language and gender recognition models, and utilizes feasible acoustic features and an optimal model-training method. From the four different model-training approaches we designed, we evaluated the system performance by the accuracy recognition rates and found the best training method. The experimental results show that it achieves high accuracy of 85.25% and 93.2%, and indicate that it will be a feasible application for Human Computer Interfaces (HCIs).