Emotion Speech Recognition Based On Artificial Neural Network

Sentiments and emotions have deeply influenced our daily life. It has great significance to let thecomputer have the ability to deal with emotional information and to improve human-computerinteraction, along with rapid development of artificial intelligence. Speech is a major approach forhuman to express emotions, detecting emotions from speech have been the subject of many studies, alot of classifiers were developed for emotion speech recognition to make the communicationbetween human and computer more convenient, natural and comfortable. Benefits from otherresearchers'works, the research showed in the paper has these contents:1. A sequence of process named pre-emphasis, framing and endpoint detection has been put onthe utterances before feature extraction..2. Feature analysis and extraction. Some statistical features, such as the duration and themaximum, minimum and mean value of short-time energy, zero-cross rate, pitch and formants, havebeen used for emotion recognition..3. After feature extraction, some feature analysis were putted on the feature space: featurereduction were taken and outliers were deleted to prepare for the next process.4. With the inspiration from physiology and psychology, the genetic algorithm and a methodcalled "innovation" mechanism were imported to improve BP network. After that, a combination ofthe two methods called IN-GABP were suggested for emotion speech recognition. At last, somecomparisons between these 3 algorithms proposed and other improvement algorithms have beenconsidered to test the performance of them.