The paper investigates the use of Fisher's ratio technique for feature selection in automatic speech recognition (ASR) using Mel Frequency Cepstral Coefficients (MFCCs). It analyzes whether a subset of MFCC coefficients can achieve comparable or improved classification accuracy when used with a Hidden Markov Model (HMM) algorithm, concluding that selecting eight coefficients based on high Fisher's ratio yields better accuracy than using all twelve. The results contribute to the understanding of dimensionality reduction in ASR and suggest further research into other phonetic units.