Research And Development On Mobile Speech Emotion Recognition System

As we all know, emotion plays key roles in effective teaching. But in current E-Learning and mobile learning environment, teachers and students are often separated from each other in time and space, which lacks necessary affective communication and thus causes emotional deficiency. The speech emotion recognition technology, as a research hotspot in current human computer interaction area, can be well applied to mobile learning scenarios, to realize the intelligent and humanistic M-Learning.This thesis researches and develops speech emotion recognition system in mobile learning environment. This system uses mobile devices as the platform, analyses the students'speech signals, and then identifies his/her emotional states, in real-time or in non-real-time.As the current speech emotion recognition technology is still not mature, there were not common emotional speech feature extraction algorithms. And the emotion recognition results might vary greatly in different language, different environment and for different gender. After doing lots of research and experiments, this thesis improves the existing emotional feature extraction algorithm. PCA algorithm is then used to reduce the dimensionality of feature vector and finally support vector machine is used to recognize three different learning emotions. Results show that our speech emotion recognition model has relatively high distinguish ability (95% in non-real-time, 80% in real-time). Meanwhile, we established a Chinese learning emotional corpus, to assist better Chinese speech emotion recognition research.Our system was first developed on PC and then transplanted to mobile phone. Considering the performance limitations of mobile devices, this thesis places the time-consuming and computing-demanding model training process on the PC side, and then imports the trained model into mobile phone, which improves the system efficiency greatly.