Project overview
The team is developing a Lithuanian telephone-conversation quality-assessment solution through the Norway Grants initiative. The product aims to recognize the set of words required for conversation quality assessment with approximately 90% accuracy when processing spontaneous Lithuanian phone conversations.
Key findings
Research revealed significant variations in recognition performance. Speaker characteristics substantially impact results: approximately 5% of speakers experience roughly double the error rate compared to average performance.
Call-center staff interactions show notably different complexity levels than customer conversations. Likely because, as the team observed, "the staff of the call centers speak more accurately and articulate better" than the general customer population, producing cleaner audio and a higher recognition rate.
Project details
The initiative receives funding from the 2014–2021 Norwegian Financial Mechanism and Lithuania's State Budget (project code: LT07-1-EIM-K02-019). Implementation partners include International Development Norway AS. The project aligns with EEA-Norway Financial Mechanisms objectives to advance economic development and international cooperation.
Editor's note: The original post was published in both English and Lithuanian. The Lithuanian version covers identical project information — contact info@oxus.ai if you need the full Lithuanian text.