• Rohith Gowtham Kodali
• Durga Prasad Manukonda
• Rajaraman Sundararajan

This paper deals with call analytics in Dravidian languages with continuous speech recognition model using Kaldi, Scikit-learn and spaCy. An important part of the speech analytics procedure is to extract the main problem statement along with its answer key and to know the customers opinion about the respective social problems. For speech recognition, we created a standard MFCC for feature extraction for HMM models to align the text and for text analytics. SpaCy NER was used for recognizing the problem and answer keys. For question and answer alignment, Scikit-learn SVM with spaCy model with some user-defined rules were adopted. Sentiment analysis is the second major part of this study and it gives positive-negative calls list by analyzing each sentence of the caller and the responder. Real-time Telugu political calls were collected as data in Andhra region (where Telugu is one of the spoken languages) and was implemented successfully. The goal is to get insights from voice calls and easily convert voice calls into text and analytics from the text for popular languages like Telugu using Multimodal.

Text Analytics, Speech Analytics, Acoustic Modeling, HMM-GMM, LVCSR (Large Vocabulary Continuous Speech Recognition), MFCC, DNN-HMM, Kaldi, IPA, NLP, Multimodal

"Speech and Text Based Analytics in Telugu Language", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.6, Issue 3, page no.238-244, March-2019, Available :http://www.jetir.org/papers/JETIRAR06054.pdf

"Speech and Text Based Analytics in Telugu Language", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.6, Issue 3, page no. pp238-244, March-2019, Available at : http://www.jetir.org/papers/JETIRAR06054.pdf