Study And Implementation Of Speech Emotion Recognition Based On DSP

Emotion plays an important role in the human perception, decision-making processing. The study of emotional intelligence has existed in psychology and cognitive science in a long time, with the development of information technology, human-computer interaction capabilities of the continually increasing demands from, emotional information processing which can improve human-computer interaction ability has become an important issue. Speech is an important means in human communication, which is the most convenient, the most basic and the most direct method in information exchanging; it contains abundant emotional information as facial expression. The purpose of emotion recognition is to identify emotional information from speech, and in order to identify these feelings, it is necessary to classify emotion. Emotion is studied in this paper, which is classified into happy, anger, surprise, sadness, fear and calm.A small scale mandarin emotional database is built by recorded, and it was taken as research object in this paper. Before emotion analysis and recognition, speech signal is digitalized and framed, making it into short-time stationary digital signal in order to be processed by computer. Aimed to the shortage of endpoint detection method, a continuous speech endpoint detection algorithm based on improved energy-zero is proposed, it obtains good detection effect in experiment.Based on the analysis on time characteristics, amplitude characteristics, fundamental frequency characteristics and formant characteristics, short-time energy, fundamental variation rate, maximum amplitude, short-time average zero-crossing rate, fundamental frequency average, first, second, third formant frequency are selected as characteristic parameters. A fundamental frequency calculation method based on self-correlation and a formant frequency calculation method based on liner prediction are shown, determine emotional contribution degree of every parameters through contribution analysis, build emotional recognition operators, and recognize speech emotion using weighted Euclidean distance. The statistical experimental results show that recognition rate of every emotion is over 60%.Using DSP experimental system, do emotion recognition experiment, the results of experiment show that each type of emotion recognition rate is more than 60%.