Speech Emotion Recognition using Attention-based LSTM-Network with Residual Connection
Researchers have developed a new, lightweight architecture called ResLSTM-SA for speech emotion recognition. This model integrates residual connections and soft attention within an LSTM framework, significantly reducing computational and memory requirements compared to large pretrained models. Tested on the RAVDESS dataset, ResLSTM-SA achieved a UAR of 0.6517 with only 46.8k parameters, making it suitable for deployment on edge devices and real-time voice assistants. AI
IMPACT Enables more efficient deployment of speech emotion recognition on edge devices and real-time assistants.