Speech recognition has been an active field of research in the last few decades since it facilitates better human–computer interaction. Native language automatic speech recognition (ASR) systems are still underdeveloped. Punjabi ASR systems are in their infancy stage because most research has been conducted only on adult speech systems; however, less work has been performed on Punjabi children’s ASR systems. This research aimed to build a prosodic feature-based automatic children speech recognition system using discriminative modeling techniques. The corpus of Punjabi children’s speech has various runtime challenges, such as acoustic variations with varying speakers’ ages. Efforts were made to implement out-domain data augmentation to overcome such issues using Tacotron-based text to a speech synthesizer. The prosodic features were extracted from Punjabi children’s speech corpus, then particular prosodic features were coupled with Mel Frequency Cepstral Coefficient (MFCC) features before being submitted to an ASR framework. The system modeling process investigated various approaches, which included Maximum Mutual Information (MMI), Boosted Maximum Mutual Information (bMMI), and feature-based Maximum Mutual Information (fMMI). The out-domain data augmentation was performed to enhance the corpus. After that, prosodic features were also extracted from the extended corpus, and experiments were conducted on both individual and integrated prosodic-based acoustic features. It was observed that the fMMI technique exhibited 20% to 25% relative improvement in word error rate compared with MMI and bMMI techniques. Further, it was enhanced using an augmented dataset and hybrid front-end features (MFCC + POV + Fo + Voice quality) with a relative improvement of 13% compared with the earlier baseline system.

Tonal Languages Automatic Speech recognition (ASR) systems are an area of interest for researchers as it is a challenging research domain. However, the main objective is to get an ASR system that has high performance whether the language used is tonal or non-tonal. But tonal languages always have degraded results as compared to non-tonal languages. Punjabi is a tonal language and tonality has led to many challenges in the designing of the Punjabi ASR system. It is an important research area to develop a Punjabi ASR system that is able to tackle the tonality of the language. In this paper, tonal and non-tonal classification is done to extract the prosodic features, which in turn enhance the word recognition rate for tonal languages. A prosodic feature has pitch related features, which can effectively understand the tone of the word and has high accuracy in recognition of words. Extracted prosodic features are fed to the ASR system individually and then later in combinations.. Results of prosodic feature's combination with Mel Frequency Cepstral Coefficients (MFCC) are compared with the MFCC based ASR results. The performance of the system is analyzed and it is observed that WER is reduced by implementing prosodic features with Relative Improvement (RI) of14%.The results are compared with the baseline result and the performance of the system is analyzed.

Water quality is a critical factor in ensuring public health and environmental sustainability. Water quality assessment methods, especially concerning their efficiency and accuracy, have become more important with the overwhelming demands on the application of artificial intelligence (AI) as emerging transformative tools. Its accurate definition helps identify health risks, optimize resource consumption, and feed sustainable practices. This study applies machine learning (ML) models to classify water quality using an integrated dataset from Telangana, India. Initial experimentation was done on individual classifiers, which include Decision Tree (DT), Logistic Regression (LR), and Support Vector Machine (SVM). Thereafter, an experiment was conducted by merging these classifiers into a voting-based ensemble model and evaluating it using both soft and hard voting strategies to improve predictive performance. It was found that soft voting ensemble resulted in the highest classification accuracy of 96.39%, with strong performance across the other metrics as well, namely, Precision 96.49%, Recall 96.39%, and F1 score 96.41%, The hard voting model and individual base classifiers were also competitive at 95.28% (hard voting), 92.08% for LR, 95% for DT, and 95.66% by using SVM. The soft voting ensemble model, therefore, shows a relative improvement of 1.46% in accuracy over the best-performing base learner and a 27.8% reduction in its error rate. It is, therefore, confirmed that ensemble learning soft voting improves the reliability of water quality classification as well as accuracy, thus providing a strong platform for future environmental monitoring systems.