Convolutional Gated Recurrent Units (CGRU) for Emotion Recognition in Odia Language

India’s enormous dialectal diversity motivates researchers to develop efficient speech emotion databases, speech features, and emotion recognition systems. Odia has a long literary history and rich dialectal variations; proper tools and assets are lacking for getting promising results for emotion analysis tasks. This paper represents a detailed study that includes the creation and evaluation of an Odia speech emotional database along with the design and testing of the accuracy label of the model considered for the emotion recognition basis. The prime objective of our work is to develop a model using Concatenated Convolution Neural Network and Gated Recurrent Unit (CGRU) for speech emotion recognition adopting prosodic and spectral features of a speech signal. Our experiments show that CGRU gives approximately 5.36% and 6.52% better results when compared to CNN and GRU for both the Odia dataset and the benchmark RAVDESS dataset. We also demonstrate that our proposed method outperforms the state-of-art methods on the RAVDESS dataset. From this experimental study we also observed that CGRU perform faster than baseline model (CNN and GRU) thus making it well-suited for use in real-time applications.