Speech recognition with combined MFCC, MODGDF and ZCPA features extraction techniques using NTN and MNTN conventional classifiers for Telugu language

Abstract

The automatic speech recognition systems are designed for human–computer interaction in trouble-free mode. Speech recognition has vast applications. Text to speech and speech to text transformations are mostly used segments in ASR. Perfect speech recognition is done by choosing proper extraction and classification techniques with respect to slang and pitch of the language. Telugu is a south Indian language which has around 120 million speakers. There are various feature extraction techniques such as LPC, MFCC, MODGDF, RASTA, DTW and ZCPA. This paper deals with the combined techniques and its comparison with individual techniques. The rate of features extracted in joint extraction techniques gives promising results comparatively with individual technique. Techniques MFCC, MODGDF and ZCPA are combined, and joint features are extracted. The next stage is to classify the features where selected technique using neural networks. Features are classified by NTN and MNTN classifiers for speaker-dependent recognition and presented by using closed and open sets. The MNTN is evaluated for several speaker recognition experiments. These include closed- and open-set speaker identification and speaker verification. The MNTN is found to perform better than NTN classifier. Speech recognition rate for Telugu language by combine MFCC, MODGDF and ZCPA extraction techniques using NTN and MNTN classification techniques are compared with excellent results.