MUMBAI, India, Sept. 14 -- Intellectual Property India has published a patent application (202641107583 A) filed by Koneru Lakshmaiah Education Foundation on September 07, 2026, for Ai-Based Adaptive English Language Learning And Pronunciation Assessment System.
Inventor includes Dr. K. B. Glory.
The application for the patent was published on September 11, 2026, under issue no. 37/2026.
Abstract: ABSTRACT OF THE INVENTION: The present invention relates to an Artificial Intelligence based adaptive English language learning and pronunciation assessment system designed to overcome the limitations of conventional computer-assisted language learning tools and classroom instruction. The system comprises a speech acquisition module, an acoustic feature extraction engine, a hybrid convolutional–transformer neural network for phoneme recognition and forced alignment, a multi- dimensional scoring module evaluating segmental accuracy, stress, intonation and fluency, an error classification engine, an adaptive curriculum engine driven by Bayesian knowledge tracing, a large-language-model based feedback generator, and a persistent encrypted learner profile database. In operation, a learner’s spoken utterance is pre-processed and converted into acoustic features that are analysed by an L1-aware deep neural model fine-tuned on both native and non-native English speech, with particular emphasis on Indian first-language backgrounds. The model produces phoneme-level posteriors and alignment timestamps. These outputs are compared against a canonical pronunciation lexicon to generate detailed scores and an error taxonomy comprising substitutions, insertions, deletions and prosodic deviations. Concurrently, linguistic content may be evaluated for lexical and grammatical appropriateness. A Bayesian knowledge-tracing algorithm maintains a probability of mastery for each phonetic skill. Based on the current mastery vector and historical performance, the adaptive curriculum engine selects optimally challenging exercises, ranging from minimal-pair drills to free speaking tasks. Feedback is delivered in visual, auditory and natural-language forms; the natural- language component is produced by conditioning a large language model on the precise error diagnosis and the learner’s proficiency level, thereby providing pedagogically meaningful explanations and practice recommendations. All interaction data are securely stored to refine both individual learner models and global acoustic models. The architecture supports mobile, web and institutional learning-management-system deployments, with optional on-device inference for privacy. A representative case study demonstrates measurable improvement: a Telugu L1 learner reduced critical phoneme error rates from 45–62 % to 15–22 % and raised overall accuracy from 78 % to 89 % within four days of adaptive practice. The invention thus furnishes a closed-loop, scalable and highly personalised solution for English pronunciation training that addresses the rigidity, limited diagnostic depth and lack of adaptive sequencing characteristic of prior art systems.
Disclaimer: Curated by HT Syndication.
