Power Industry Telephone Telephone Customer Service Voice Emotion Recognitio

Speech emotion recognition as an indispensable part of the field of artificial intelligence, has important application value in speech processing and affective computing etc.. Speech as the emotional exchange of the people the most convenient and most common way, the speech emotion recognition has become a research direction of the current popular. The speech emotion recognition is applied to the power industry call voice, there can be a large number of idle customer service voice to solve the problem of power company is very good, also can provide good customer feedback for the electric power company. Because the power customer voice itself has its own characteristics, this paper excavates its phonetic features, into the speech emotion recognition to get better effect. The following are several aspects of the research results of this paper:(1)Speech corpus contains three sources, namely, the power company to provide, online collection and recording. By listening to a speech corpus selection and manual annotation (emotion categories) process, construction of the electric power industry customer service telephone speech corpus. Then the corpus is evaluated and compared with other commonly used emotional speech corpus, and it is concluded that this corpus is a qualified and medium scale speech corpus. This corpus provides a corpus based support for the speech emotion recognition of the telephone customer service in power industry.(2)In order to effectively utilize the local features of telephone customer service speech in power industry, a speech emotion recognition method is proposed, which is based on the local features of emotion words and the global features of speech sentences. The method dependent on acoustic feature library of electric power customer service speech emotion dictionary, and extract sentence speech is included in the emotional words and sentiment words density of local features, and integration with the global acoustic characteristics, again through the machine learning algorithm modeling and emotion recognition of speech. Experimental results show that integration of electric power customer service emotional words local features and global features of the speech emotion recognition method can achieve better effect, the local characteristics of the emotional words introduced to effectively improve the electric power customer service speech emotion recognition accuracy.(3)Power industry telephone customer service speech has its own industry characteristics, for the in-depth mining industry knowledge, this chapter puts forward the integration of power industry telephone customer service speech semantic features for speech emotion recognition. The semantic feature of the power customer service is mainly refers to the words which are closely related with the electric power industry. These words express the speaker’s meaning and emotion to a large extent. The extraction of semantic features depends on speech recognition, which can be converted into words, and then the words are matched with the semantic database, and the semantic features are obtained by computing the weights of the semantic words in the sentence. Finally, through the experiment of fusing the semantic features of customer service, it is concluded that the power customer service speech emotion recognition is an effective method.(4)Design and implement the prototype system of telephone customer service speech emotion recognition in power industry. The prototype system includes voice input module, the speech features processing module, emotion model training module, module of emotion recognition, recognition result output module, special in the speech processing module, contains the local feature sub module and semantic features of sub module. The prototype system provides an important support for the telephone customer service in the power industry.