Dialectal Variations of Isolated Word Recognition

Abstract

Speech can be considered as most important aspect of communication among living creatures. A lot of work has been done in the past in the area of speech processing but it has a wide variety of applications such as speech recognition, speaker identification, speech synthesis, machine translation, information retrieval system and others. The objective of this paper is to discuss the performance of dialectal variations of isolated word recognition. Hidden Markov Model (HMM) technique is used for implementation of isolated speech recognition system. MFCC technique is used for feature extraction. Speech corpus consists of 125 isolated words, spoken by 100 speakers i.e. 30 males and 30 female speakers of Malwi dialect, 10 males and 10 female speakers of Majhi dialect and 10 males and 10 female speakers of Doabi dialect. System performance is tested in computer lab environment by male speakers and female speakers. Speech recognition system is also tested in two modes i.e. by speakers involving in both training and testing phase and by speakers involving in testing only. In the first mode, Speech recognition accuracy is 82.56% by Malwi dialect male speakers, 78.80% by Malwi dialect female speakers and 80.68% by Malwi dialect mixed (males and female) speakers. Speech recognition accuracy is 68.80% by Majhi dialect male speakers, 87.28% by Majhi dialect female speakers and 78.04% by Majhi dialect mixed (males and female) speakers. Speech recognition accuracy is 86.40% by Doabi dialect male speakers, 81.68% by Doabi dialect female speakers and 84.04% by Doabi dialect mixed (males and female) speakers. In the second mode, Speech recognition accuracy is 83.12% by Malwi dialect male speakers , 79.52% by Malwi dialect female speakers and 81.32% by Malwi dialect mixed (males and female) speakers. Speech recognition accuracy is 63.68% by Majhi dialect male speakers, 83.60% by Majhi dialect female speakers and 73.64% by Majhi dialect mixed (males and female) speakers. Speech recognition accuracy is 83.28% by Doabi dialect male speakers, 79.60% by Doabi dialect female speakers and 81.44% by Doabi dialect mixed (males and female) speakers.