Hybrid Features for Neural Network-Based Bangla ASR Incorporrating Velocity Coefficients (?)

This paper presents a Neural Network-based Bangla phoneme recognition method for Automatic Speech Recognition (ASR). The method consists of three stages: at first stage, a multilayer neural network (MLN) converts acoustic features, mel frequency cepstral coefficients (MFCCs), into phoneme probabilities, where the second stage computes velocity (Δ) coefficients from the phoneme probabilities by using three point linear regression (LR). Finally, the phoneme probabilities, velocity (Δ) coefficients and the input MFCCs, combined as hybrid features, are fed into a hidden Markov model (HMM) based classifier to obtain more accurate phoneme strings. From the experiments on Bangla speech corpus prepared by us, it is observed that the proposed method provides higher phoneme recognition performance than the existing method. Moreover, it requires a fewer mixture components in the HMMs.