Prosodic Feature-Based Discriminatively Trained Low Resource Speech Recognition System

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.

Enhanced Content-Based Image Retrieval Using Information Oriented Angle-Based Local Tri-Directional Weber Patterns

In diverse computer applications, the analysis of image content plays a key role. This image content might be either textual (like text appearing in the images) or visual (like shape, color, texture). These two image contents consist of image’s basic features and therefore turn out to be as the major advantage for any of the implementation. Many of the art models are based on the visual search or annotated text for Content-Based Image Retrieval (CBIR) models. There is more demand toward multitasking, a new method needs to be introduced with the combination of both textual and visual features. This paper plans to develop the intelligent CBIR system for the collection of different benchmark texture datasets. Here, a new descriptor named Information Oriented Angle-based Local Tri-directional Weber Patterns (IOA-LTriWPs) is adopted. The pattern is operated not only based on tri-direction and eight neighborhood pixels but also based on four angles [Formula: see text], [Formula: see text], [Formula: see text], and [Formula: see text]. Once the patterns concerning tri-direction, eight neighborhood pixels, and four angles are taken, the best patterns are selected based on maximum mutual information. Moreover, the histogram computation of the patterns provides the final feature vector, from which the new weighted feature extraction is performed. As a new contribution, the novel weight function is optimized by the Improved MVO on random basis (IMVO-RB), in such a way that the precision and recall of the retrieved image is high. Further, the proposed model has used the logarithmic similarity called Mean Square Logarithmic Error (MSLE) between the features of the query image and trained images for retrieving the concerned images. The analyses on diverse texture image datasets have validated the accuracy and efficiency of the developed pattern over existing.