ANN Based Recognition of the Dimasa Tribal Language of North East India

Speech is a fascinating phenomenon in the human body. Speech differs from one person to another. Most likely, no two people have the same properties of speech like pitch, formant frequencies, duration, etc. Also, different people speak different languages. All these stated properties make speech/language a non-linear and non-stationary signal, which is challenging to implement for a speech/language recognition system. Here we have tried to design the language recognition of the Dimasa language by first implementing a single-word recognition system. This Dimasa language is a Sino-Tibetan language. It is tonal in nature and spoken by a few local tribes in North Eastern part of India. It is predominant in the Districts of Dima Hasao, Karbi Anglong, Nagaon, and Cachar of Assam. Along with the Dimapur District and Jiribam region of Nagaland and Manipur states respectively. Different techniques of Artificial Neural Networks (ANN) were used to obtain different results. The approach is based on feature extraction techniques Mel-Frequency Cepstrum Coefficients (MFCC). In this work, the popular Levenberg–Marquardt training algorithm, Scaled Conjugate Training Algorithm, and Gradient Descent Propagation Training algorithm were used to train for the pattern recognition neural network used to identify the language and speech.