The objective of the proposed research is to establish an unified framework for multilingual ASR based on universal speech attribute modeling. Specifically, instead of the traditional end-to-end ASR systems that rely on collecting a large amount of speech and text data and learning the model parameters without the need to use detailed knowledge about a target language, our proposed approach is guaranteed to be knowledge-rich. Our preliminary work on language-universal approach shows that by modeling the universal speech attributes, significant improvement can be achieved for both phoneme and spoken keyword recognition. Building on these findings, the research will now focus on expanding the framework to handle continuous speech recognition, seeking to address the limitations of current multilingual ASR systems.