Verbalystic: An AI-Powered Multi-Dimensional Spoken Communication Assessment System

Abstract

Verbal communication ability is an essential skill base, but this is the area where learners fail to locate the tools that enable them to have on-going, objective, and multi- dimensional feedback. Present day interventions in Computer Assisted Pronunciation Training (CAPT) and Automatic Speech Recognition (ASR) are often afflicted with severe shortcomings: reliance on costly proprietary APIs, narrow focus primarily on phoneme accuracy, and absence of real-time diagnostic analysis. In this paper, we present Verbalystic, an innovative AI-powered spoken communication assessment system designed to address these obstacles. Verbalystic utilizes Web Audio API to capture and record audio in real-time with low latency, and uses the Vosk ASR toolkit to transcribe audio with high accuracy at low cost. The main advantage of the system is its multiaspect analysis pipeline, where metrics of speech such as rate, pitch, volume, transparency, frequency of filler words and repetition are computed. It creates refined, granular feedback, which provides a holistic pronunciation score, contextual grammar correction, assessment of semantic coherence, and an AI-enhanced transcript. Built on a scalable and cost-effective stack of JavaScript, FastAPI, and PostgreSQL, Verbalystic is highly accessible. Furthermore, it uses powerful interaction elements such as gamification (streaks and badges) and in-depth progress dashboards to maintain long- term engagement and learning. Verbalystic has shown potential to increase the level of error awareness and improve communicative habits, establishing a new standard for accessible and thorough spoken language training.