Dysarthria, a speech disorder stemming from difficulties in controlling the relevant muscles, presents formidable challenges to effective communication. This study proposes an Automatic Speech Recognition (ASR) system that uses the encoder of Whisper model to extract features from audio files containing a single utterance of word and passes it to the classifier model for identification. This transfer learning approach builds a speaker-independent dysarthric ASR system by employing features obtained from the Whisper encoder module into three classifier models - LSTM, Bi-LSTM, and Bi-GRU. The whisper feature set achieved high accuracy of 71.55% when used with BiGRU model, surpassing relatively the best average accuracy achieved by speaker-adaptive ASR system of Speech Vision. Robustness of these models was analyzed against AWGN and babble noise at various SNR levels. The performance degradation is significantly less for proposed Whisper features than state-of-the-art wav2vec2.0 model.