Conceiving Entirely New Language for Speech Analytics

In the ever-evolving landscape of global business solutions, last years marked pivotal moments for us. At one of which, a message arrived from Eleveo, a respected global partner renowned for its expertise in Workforce Engagement Management (WEM) solutions. This message carried within it a challenge that would ignite a journey of innovation and collaboration. They needed Speech Analytics solution in the Latvian language, which means conceiving an entirely new language from the ground up.

Embarking on the Challenge

Eleveo's client in Latvia yearned for a high speech recognition accuracy, a critical component for analysis tools. However, the existing local Automatic Speech Recognition (ASR) vendor was struggling to achieve a mere 60% accuracy in recognizing Latvian speech. Despite Latvian not being among our supported languages, we embraced the challenge.

The project demanded expertise, dedication, and a fresh perspective. Sestek and its local partner, TET Latvia, a distinguished mobile operator with native linguistic proficiency, joined forces. The journey of collaboration and determination began in June 2021.

A Remarkable Achievement

The collaborative project was not without its challenges, but through the combined skills and efforts of the teams involved, a remarkable milestone was achieved. By the end of the year, Sestek and its partners had developed a Latvian language model that achieved an astonishing accuracy level of 96% in Speech Analytics. This achievement surpassed expectations and met the client's timeline but also showcased Sestek's exceptional ability to craft entirely new languages from scratch.

This accomplishment stood as a testament to the power of innovation driven by collaboration. It highlighted how a challenge could be transformed into an opportunity through the synergy of expertise and dedication. The journey showcased the convergence of technological know-how and linguistic proficiency, ultimately revolutionizing the field of Speech Analytics.

Our Secret Sauce: End-to-End Model

The collaboration between Eleveo, Sestek, and TET Latvia overcame the linguistic barriers; the joint efforts not only fulfilled a business need but also led to a big success. The main pillars of this success were the coordination between teams and 2 decades of expertise of Sestek in voice technologies.

The collaborative project commenced its journey in June 2021, and through concerted efforts, we achieved a remarkable milestone. By the culmination of the year, we successfully engineered a Latvian language model boasting an impressive accuracy level of approximately 96%. 

To briefly explain our technical point of view; for more accurate speech recognition, we don’t break down words into parts but analyze them within context. This is called End-to-end Model. We train and use a single model for the speech recognition process. Using Deep Neural Network (DNN) and Recurrent Neural Network (RNN) methods, our SR engine can understand complex issues like sound waveforms, language structures, and features of speech. These AI algorithms provide us with market-leading accuracy and flexibility.  

Author: İlhan Karakoc, Regional Sales Director- UK&EU

For more information pls visit our Speech Recognition page : https://www.sestek.com/speech-recognition-knovvu