Bangla Speech Emotion Recognition Using Bidirectional Gated Recurrent Unit and Two Dimensional Convolution Neural Network

In the field of artificial intelligence, speech recognition is regarded as a growing and progressive research area. The speech signal is advanced with paralinguistic and semantic data. A prominent and courageous example of paralinguistic information partially expressed through speech is emotion. Human-machine collaboration will arrive at another level when machines comprehend paralinguistic data, like inclination. This study utilizes the Bengali language and AI to deal with perceived feelings in discourse. We collected a Bengali speech feeling corpus named 'SUBESCO'. Bi-directional Gated Recurrent Unit (BiGRU) and Two Dimensional Convolution Neural Network (2DCNN) were utilized for the arrangement task, and as elements, we utilized Mel Spectrogram. Our Bi-GRU & 2DCNN model gave 95.64% exactness on expanded sound documents. We additionally approved RAVDESS dataset by applying CNN, BiGRU, GRU, and BLSTM, which are notable AI calculations. The Mel- Spectrogram highlights of the sound were also involved here. In any case, our proposed BiGRU & 2DCNN approach gave preferable execution over different techniques we applied to our dataset. These outcomes prove that our discourse feeling corpus can be utilized effectively for research work in the field of discourse feeling acknowledgment.

• Supervision