Audio Insight: Audio Stream Context Recognition, Intent and Entity Classification using SNIPS NLU and SMER Named Entity Recognizer
Tushar Agarwal, Bhavesh Singh, Shweta Sinha, Priyanka Makkar · 2024
Understanding of conversational context by software and machines is of utmost importance for the growing need for Artificial Intelligence interfaces, facilitating effective human-machine communication. This need grows exponentially when dealing with the medical domain, having machines respond faster and accurately to medical use cases is crucial. The research, hence, rigorously analyzes the performance of the SNIPS NLU engine along with our novel SMER (SciSpacy and Med7 based NER) Medical Named Entity Recognizer. The research then furnishes a mechanism to understand patient complaints and provides prompt results based on the retrieved information. Specifically, the study involves using the SNIPS NLU for context classification and SMER for information retrieval in an ensemble approach to provide an Intent classification (NLU) accuracy of 94% and Entity accuracy of (SMER) 86.95%, thus posing as a pivotal benchmark in advancing machine proficiency for finding context from audios. The proposed mechanism, thereafter, can be used in situations involving booking medicines, procedures, appointments, and more.