This paper aims to analyse how innovative Artificial Intelligence (AI) systems for non-standard speech recognition may revolutionize Augmentative Alternative Communication (AAC) technology for people with severe speech impairments. The AI-based system is personalized for each person's unique speech production and offers a real step forward in improving the efficiency of AAC. With impressive enhancements in recognizing non-standard natural language supported by Machine Learning and Deep Learning algorithms, AI is offering a turning point for personalized and customized Augmentative Alternative Communication (AAC). However, there is a need of understanding the contextual needs of the user which enhances the experience while using AAC. To address this, a systematic review has been done to identify existing applications and the technologies behind them. Also, challenges are explored that can lead to the future directions of the research work. This paper presents the results of the systematic review, after filtering 62 journal articles that are more relevant to the context from a pool of 1088 papers, in the perspective of the existing AI technologies in speech generation, speech reconstruction and speech generation. AI based AAC applications are comprehended in term of input/output, cost, infrastructure and user background. The paper also highlights the future research directions in this domain by identifying the research gaps.