부산대학교 도서관

Features play a vital role in Automatic Speech Recognition (ASR). They also aect the performance of speech recognizers in all environments to an extent. There are many types of features used in automatic speech recognition. Each feature type has its own signicance depending upon its unique characteristics. LPC (Linear Predictive Coding), LPCC (Linear Predictive Cepstral Coecients), Mel Frequency Cepstral Coecients (MFCC), Perceptual Linear Prediction (PLP) and Relative spectral Analysis (RASTA) are some of the features that are generally used. Better results can be obtained by combining coecients of dierent features. For instance, MFCC and PLP show better results when they are jointly used to recognize a speech signal. This paper aims to compare the performance of dierent features using ANN for same number of speech datasets. The result shows that proposed technique increases the recognition rate by 8.33% as compared to MFCC alone.