Intelligent Automation Platform for Information Recording in Telephonic and Spontaneous Environments

The advancement of artificial intelligence and process automation is undeniable. Within this paradigm, data is considered “the new oil,” as it powers AI and automation systems. Data-driven systems are used for a wide range of tasks: quality recording, defect traceability, understanding heterogeneous business realities, triggering and automating processes, knowledge discovery, and more. Ultimately, data is a unit containing information, and DEITU aims to exploit the most natural and human data available: speech. In today’s data-driven society, obtaining information from this communication channel is essential to apply data-based mechanisms in the daily operations of various companies.

This is the scenario presented by GSR and SARETEKNIKA within the context of DEITU. In their daily operations, both companies handle processes that involve speech, whether for managing incidents or monitoring conversations between family members and caregivers. Although speech recognition technology has advanced significantly in recent years, the DEITU project faces serious challenges, such as spontaneity and the use of the telephone channel.

On top of this, extracting semantic information from unstructured text using natural language understanding algorithms, while disambiguating and structuring interactions through conversational assistants in spontaneous and natural interactions, is a significant and interesting challenge. Furthermore, to reduce the operational cost of maintaining and managing these systems, the paradigm of Active Learning is introduced. In this approach, the system itself identifies the data segments that generate the most uncertainty in information extraction, optimizing effort and increasing the efficiency and effectiveness of human operators given the large volume of data to analyze.

The opportunity presented by DEITU is clear: the project will exploit the telephone channel for the extraction and structuring of conversational information. Enabling this channel will extend speech recognition and natural language understanding technologies to a widely used channel, particularly among older generations, and across many application fields. Moreover, working with spontaneous conversations presents both a technological challenge and a clear opportunity: developing information extraction technologies in more informal environments than guided conversations will allow the deployment of natural language understanding modules, assistants, and speech recognizers in market niches where the technology had previously been unable to operate due to lack of robustness and negative impact on user experience when automating these tasks. Finally, incorporating active learning for intelligent post-production monitoring and information management will reduce the operational cost of maintaining and improving information capture systems, allowing technology and innovation to reach the industrial market with less friction.