Building an ASR System for Indian (Punjabi) language and its evaluation for Malwa and Majha dialect: Preliminary Results

Automatic Speech recognition (ASR), also referred to as voice recognition or speech-to-text, is one of the most developing field of Natural Language Processing (NLP) that recognizes speech. Speech recognition allowing the human voice to serve as the main interface between the human and the computer. Plenty of progress is to be done in the field of speech recognition for various popular languages and adult users with good recognition rates. But in case of Indian regional languages Punjabi, Telugu, Tamil etc, speech recognition is still at the infant level under disparate acoustic conditions. The objective of this research work is to demonstrate the effectiveness of the pitch acoustic features for improving the recognition performances of the ASR system for different Punjabi dialects. When our developed dialect-based Punjabi ASR system was evaluated for Malwa and Majha dialect speakers, the findings showed WER of 23.25 % and 25.91 %, respectively.