Deep neural networks(DNN) and Gaussian mixture model(GMM) has recently achieved significant performance gain and better efficiency in spoken language recognition(SLR) which identifies the language being spoken. But in our system we propose language translation and transcription of input signal (human voice) which produces output in text form of a standard language depending on the user’s application area. Senone posteriors used in context- dependent deep neural network (CD- DNN) which recognises the language spoken by people irrespective of pronunciation. Wavelet (HAAR) transform is applied on voice signal to obtain features of voice. Feed forward network issued to pass the input signal to hidden layer of NN network and obtain the primitive function of node. Computational speed of 1-3 sec is achieved, Significant improvement in gain and efficiency of (65-73%) is obtained compared to Gaussian mixture model (GMM).
