Abstract:
Improving user experience is one of the most important goals of power marketing. Power customer service is the first-line system to contact customers in power marketing, ...Show MoreMetadata
Abstract:
Improving user experience is one of the most important goals of power marketing. Power customer service is the first-line system to contact customers in power marketing, and customer emotion is an important indicator to describe customer satisfaction. Therefore, it is a very valuable task for power marketing to analyze the user's feelings from the speech data of power customer service. However, the speech emotion recognition based on power customer service is vulnerable to the interference of emotion independent factors, such as speaker differences, environmental noise, voice quality and so on. And the data has unbalanced sample classes. In this paper, we calculate the spectrogram of speech and its first-order and second-order difference, stack the three as the input of neural network to reduce the influence of emotion independent factors; we use CNN and LSTM to extract speech data features, and add attention mechanism to make the model focus on the time-frequency region related to emotion. For the problem of unbalanced sample classes, we add Focal Loss to reduce the weight of samples which are easy to classify in loss. The results of experiment show that the recognition accuracy of our model is 92.60% of weighted accuracy and 92.02% of unweighted accuracy, which is significantly improved compared with the traditional DNN-ELM method.
Date of Conference: 10-13 December 2021
Date Added to IEEE Xplore: 17 January 2022
ISBN Information:
Funding Agency:
Keywords assist with retrieval of results and provide a means to discovering other relevant content. Learn more.
- IEEE Keywords
- Index Terms
- Customer Service ,
- Emotion Recognition ,
- Speech Emotion Recognition ,
- Neural Network ,
- Convolutional Neural Network ,
- Long Short-term Memory ,
- Attention Mechanism ,
- Network Input ,
- Environmental Noise ,
- Focal Loss ,
- Input Of Neural Network ,
- First-order Difference ,
- Second-order Difference ,
- Voice Quality ,
- Speech Data ,
- Improve User Experience ,
- Unbalanced Classes ,
- User Feels ,
- Model Performance ,
- Negative Emotions ,
- Attention Layer ,
- Convolutional Layers ,
- Real Class ,
- Deep Neural Network ,
- Linear Layer ,
- ReLU Activation Function ,
- Softmax Layer ,
- Speech Segments
- Author Keywords
Keywords assist with retrieval of results and provide a means to discovering other relevant content. Learn more.
- IEEE Keywords
- Index Terms
- Customer Service ,
- Emotion Recognition ,
- Speech Emotion Recognition ,
- Neural Network ,
- Convolutional Neural Network ,
- Long Short-term Memory ,
- Attention Mechanism ,
- Network Input ,
- Environmental Noise ,
- Focal Loss ,
- Input Of Neural Network ,
- First-order Difference ,
- Second-order Difference ,
- Voice Quality ,
- Speech Data ,
- Improve User Experience ,
- Unbalanced Classes ,
- User Feels ,
- Model Performance ,
- Negative Emotions ,
- Attention Layer ,
- Convolutional Layers ,
- Real Class ,
- Deep Neural Network ,
- Linear Layer ,
- ReLU Activation Function ,
- Softmax Layer ,
- Speech Segments
- Author Keywords