Speech-to-text

Reading time: 2 minutes

In collaboration with the Data Lab of the SVB and the TOP4 (Talent Development Program), a Design Sprint was carried out on the topic of telephone conversations and gaining insight into customer satisfaction. The speech-to-text technique can be a solution for service providers because it automatically generates conversation reports and an analysis can then be released.

Insight into telephone conversations

In collaboration with the Datalab and TOP4 (Talent Development Program) of the SVB, Novum started to conduct a Design Sprint on the topic of telephone conversations and to gain insight into customer satisfaction. We did an experiment with the speech-to-text technique. During the Design Sprint, two start-ups started designing a prototype that connects to the question why customers call the SVB. Use was made of audio with simulated telephone conversations. The two prototypes give an impressive picture of how converting speech to text can be a solution to find out the reasons for a call. This can ultimately contribute to the targeted improvement of services.

The SVB is hardly clear why its customers contact us by telephone. No telephone conversations are recorded and at least a report of completed customer contact is made. This of course has its reasons. However, the outcome of the sprint shows how enriching the information from telephone conversations can be for service provision - and how well this can be facilitated through speech-to-text.

Speech-to-text can be a solution for service providers because it automatically generates call reports. In addition, it gives the possibility to analyze why, when or how often customers call and how satisfied they are after the conversation. An accuracy of 70-75% applies for converting Dutch speech into text for conversations between two people. The technology requires further development when it comes to conversations between more than 2 people and conversations with a lot of background noise.

A simple yes or no answer to the question whether the customer has been helped after the telephone conversation can already lead to important insight into where improvements in services can be made. In addition to being able to analyze more easily, speech-to-text provides advanced options in categorizing conversation topics and recognizing sentiment. Words such as 'unfortunately', 'annoying' or 'fine' and 'reassured' can indicate whether the customer is satisfied or not.

The SVB has continued with the results of the sprint.

Share this article