Enhanced Systematic Structure for Data Pre-Processing and Intensifying Data Augmentation for Audio Data Processing with SS-GRNN Technique

Speech recognition technique for tribal language helps tribal children with their education. To automate speech recognition by the tribal audio dataset. This audio dataset should be cleansed, pre-processed, and intensified for better speech recognition, translation, and for better accuracy. The enhanced Systematic Structured GRNN (SS - Gaussian mixed Recurrent Neural Network) heterogeneous model pre-processed and augmented these data. The GRNN model is used to build a structured Data Cleaning, Data Processing, and Augmentation format. This structure also includes the following works: identifying noise from the audio, missing data, adding domain knowledge, transforming all audio to the same scale, and handling imbalanced data. This proposed model facilitates speech recognition to improve accuracy and adequate translation for the tribal language. The proposed SS-GRNN model produced 94.34% accuracy higher than the mADC-RNN conventional model with 84% accuracy.