Combining the Evidences of Temporal and Spectral Enhancement Techniques for Improving the Performance of Indian Language Identification System in the Presence of Time-Varying Noises

Language Identification has gained significant importance in recent years, both in research and commercial marketplaces, demanding an improvement in the ability of machines to distinguish between languages. Although methods like Gaussian Mixture Models, Hidden Markov Models, and Neural Networks are used for identifying languages, the problem of language identification in time-varying noisy environments has not been addressed effectively. In this paper, varying noise characteristics are simulated using three approaches: (i) Same SNR with different noises, (ii) Same noise with different SNRs, and (iii) Different SNRs with different noises. A comparative performance analysis of speech enhancement techniques like Minimum Mean Squared Estimation (MMSE), Spectral Subtraction (SS), and Temporal Processing (TP) is presented. Although these speech enhancement techniques independently may not yield superior performance, combining the data from all these techniques appears to develop an enhanced identification system. The language identification studies are performed using the IITKGP-MLILSC (IIT Kharagpur-Multilingual Indian Language Speech Corpus) database consisting of 27 Indian languages.