In 2021, Oxus.AI was selected by Norway Grants as the second-best AI project, which came with a pleasant bonus of 200K in grants. This success resulted in the raising of 800K in the angel round, which mostly consisted of micro VCs and market experts. Since then, Oxus was on the quest of democratizing AI speech analytics.
The project with Norway Grants (financed by the 2014-2021 Norwegian Financial Mechanism) formally started in October 2021 as a “Lithuanian telephone conversation quality assessment solution.” Its main goal was to develop an AI speech analytics solution for the Lithuanian language to analyse conversations over the phone. The trick here was that at the time there was no reliable Speech-to-Text solution that would perform consistently at, at least, 65% accuracy on telephony quality data, which put the endeavor and the team on the frontlines of the research. The initial plan was to go with a keyword spotting approach, which means that only certain words will be extracted from the audio file, without needing to transcribe the entire audio. This was supposed to solve the problem of the absence of a full speech-to-text component in the solution.
According to the Norway Grants scientific research schedule, Oxus.AI needed to achieve a 70% accuracy when compared to the manual evaluation of phone conversations. The team at Oxus.AI found out fast enough that the keyword spotting approach does not work as well as they had hoped for. This led to the decision to go with a full Speech-to-Text to collect the data from the calls for further NLU exploration. With the help of real-life data, the team managed to train the models and achieved a stunning call evaluation accuracy rate of 90%. Not only this achievement surpassed expectations, but also it came with an understanding of the possible implications for other languages as well.
This success at attaining a high level of accuracy on real-world data surely has greater implications than simply a good automatic quality assurance for Lithuanian call centers. It seems that the process can be repeated with virtually any language. So soon enough there will be an opportunity for even the smallest language to enjoy the spoils of automated QA.
The team at Oxus.AI is constantly looking for new challenges and opportunities to help customer support and sales teams thrive. We hope soon enough we will have even more insights from our research!
Oxus.AI started actualizing the project “Lithuanian telephone conversation quality assessment solution” financed by the 2014-2021 Norwegian Financial Mechanism and the State Budget of the Republic of Lithuania, project code no. LT07-1-EIM-K02-019, according to the 2014-2021 Norwegian Financial Mechanism Program for Business Development, Innovation, and SMEs in the field of information and communication technologies. More information about the “Business Development, Innovation and SMEs” program can be found here: https://www.eeagrants.lt/programos/aprasymas/program/5
