Emotion Recognition and Evaluation of Mandarin Speech Using Weighted D-KNN Classification

Emotion Recognition and Evaluation of Mandarin Speech Using Weighted D-KNN Classification Tsang-Long Pao, Yu-Te Chen, Jun-Heng Yeh, Yuan-Hao Chang Department of Computer Science and Engineering, Tatung University, Taipei tlpao@ttu.edu.tw, d8906005@mail.ttu.edu.tw, g9206026@ms2.ttu.edu.tw Abstract. In this paper, we proposed a weighted discrete K-nearest neighbor (weighted D-KNN) classification algorithm for detecting and evaluating emotion from Mandarin speech. In the experiments of the emotion recognition, Mandarin emotional speech database used contains five basic emotions, including anger, happiness, sadness, boredom and neutral, and the extracted acoustic features are Mel-Frequency Cepstral Coefficients (MFCC) and Linear Prediction Cepstral Coefficients (LPCC). The results reveal that the highest recognition rate is 79.55% obtained with weighted D-KNN optimized based on Fibonacci series. Besides, we design an emotion radar chart which can present the intensity of each emotion in our emotion evaluation system. Based on our emotion evaluation system, we implement a computer-assisted speech training system for training the hearing-impaired people to speak more naturally. 1 Introduction Recognizing emotions from speech has gained increased attention recently, because automatic emotion recognition can help people to develop and design many applications about human-machine communication. In emotion recognition, collecting corpus and selecting the suitable features and classification algorithms are the two most difficult problems. Language is the most basic and main tool for the human to communicate thoughts, convey messages and express aspiration. For the hearing-normal people, the process of speak learning is very natural. But for the hearing-impaired people, it becomes almost impossible due to no auditory input. Fortunately, the hearing-impaired people are not profoundly deaf and remain some level of hearing. Using these residual hearing and other perception, the hearing-impaired people can still communicate with other people. In real life, we can often see that hearing-impaired people converse with others by sign language, lip reading or writing. In fact, sign language has low popularity among general people. Lip-reading is just a reference because it has some limitations in Mandarin vowels. Writing is not a convenient way. So in many language training, to teach the hearing-impaired people to speak is the ultimate goal. For this reason, we want to design a computer-assisted emotional speech training system. By using this system, it can assist the hearing-impaired people to learn not only to speak correctly but also to speak naturally, just like the hearing-normal people. Besides, we also use the visual feedback in our system. Just like in many singing training system, we can see the singing score on the screen after singer has sung. Speech therapist can have no trouble to teach hearing-impaired people to speak with emotions when they communicate with people. This mechanism can let the hearing-impaired people better understand their 1 speaking state and make the whole system more complete. The emotional state of a speaker can be identified from the facial expression [1] [2] [3], speech [4] [5] [6], body language, perhaps brainwaves, and other biological features of the speaker. A combination of these features may be the way to achieve high accuracy of recognition. But they all are not …

Content truncated to 500 words