Spoken Gujarati Language Processing

Abstract

In the recognition of speech there are various types of techniques that translates human speech waves/frequency into readable words or in various different forms which is very easily understable by machines. Speech recognition is the very interesting field for researchers who works for different regional language processing which is easily understand by computers. For most popular languages like English and all these technologies reached at top level. For audio Classification and NLP purpose in market many other various types of Classification Techniques/methods are available. Here in this paper, I represent one model which is being best recognizer for Gujarati Spoken Digits. Here, my work finding the good and purposeful use of Artificial Neural Network (ANN) and Convolutional neural Network (CNN) for Spoken Gujarati Digit Dataset. Mainly CNN is used in the purpose for Image Classifier but here time – frequency graph of spoken Gujarati digits used in this for getting Wave graphs. In specific, wavelet transform is utilized in shaping the time-frequency representation because it gives superior recurrence localization for low recurrence signals such as speech. The time-frequency representation is resized to a common measurement utilizing bicubic addition and the coming about image-like representation, referred as Mel-spectrograms, is utilized for recognizing talked digits utilizing CNN. This model I use ANN and CNN for particular findings of finding audio classifications and similarity. At the end after testing and training period the findings are 99 percentage accuracy and it’s Val accuracy is 79 percentage have been gained by running this model at 100 epochs in CNN.