基于LSTM神经网络的语音情绪识别  

Speech Emotion Recognition Based on LSTM Neural Network

作　　者：辛创业 许芬[1] 

机构地区：[1]北方工业大学,北京100144

出　　处：《工业控制计算机》2020年第8期87-89,共3页Industrial Control Computer

摘　　要：随着人工智能的发展,人机交互技术在不断进步,为使人机交互更加友好,情绪识别技术被广泛关注。情绪是一个人内心的感触的体现,可以体现在面部、语音、脉搏等多方面。实验室环境中的语音识别技术取得了较好的效果,而现实场景的语言情绪识别技术仍不成熟,使用基于现实场景的CHEAVD2.0情感数据库进行实验。在对音频信息进行预处理后,进行音频特征的提取,提取了梅尔倒谱系数短时过零率、基音周期和频率等特征。为抓取音频数据在时间维上的关联性,使用长短时记忆网络的方法进行情绪识别分类任务。With the development of artificial intelligence, human-computer interaction technology is improving.In order to make human-computer interaction more friendly,emotion recognition technology is widely concerned.Emotion is the reflection of one’s inner feelings,which can be reflected in face,voice,pulse and other aspects.The speech recognition technology in the laboratory environment has achieved good results,but the audio emotion recognition technology in the real scene is still immature. In this paper,cheavd2.0 emotion database based on the real scene is used for the experiment.After preprocessing the audio information,extract the audio features,such as Mel cepstrum coefficient,short-time zero crossing rate,pitch period and frequency.In order to capture the relevance of audio data in time dimension,this paper uses the method of long-term and short-term memory network to carry out the task of emotion recognition and classification.

关 键 词：情绪识别 长短时记忆神经网络 梅尔倒谱系数 

分 类 号：TN912.34[电子电信—通信与信息系统] TP183[电子电信—信息与通信工程]