SER | 语音情绪识别 | TIM-NET_SER项目实现，以及训练自己的语音数据集，后期修改网络

大家好，今天复现的是目前语音情绪识别的SOTA论文，论文中文名称是时间建模的重要性：用于语音情感识别的新型时空情感建模方法。论文中训练的数据集有英文德语等几个语音情绪识别中常见的语音情绪数据集，以对比精度权重等效果~各数据集的情绪数量不同，可参考以下代码

CASIA_CLASS_LABELS = ("angry", "fear", "happy", "neutral", "sad", "surprise")#CASIA
EMODB_CLASS_LABELS = ("angry", "boredom", "disgust", "fear", "happy", "neutral", "sad")#EMODB
SAVEE_CLASS_LABELS = ("angry","disgust", "fear", "happy", "neutral", "sad", "surprise")#SAVEE
RAVDE_CLASS_LABELS = ("angry", "calm", "disgust", "fear", "happy", "neutral","sad","surprise")#rav
IEMOCAP_CLASS_LABELS = ("angry", "happy", "neutral", "sad")#iemocap
EMOVO_CLASS_LABELS = (

SER | 语音情绪识别 | TIM-NET_SER项目实现，以及训练自己的语音数据集，后期修改网络

猜你喜欢