Analysis of Layer-Wise Training in Direct Speech to Speech Translation Using BI-LSTM