We were once again in a nursing home with Navel and tested the latest prototype. Thanks to the Pflegezentrum Eichenau, a home of the Innere Mission München. We had a total of 7 test subjects: male and female nursing home residents who are still relatively fit cognitively. Some, however, have limitations in hearing and vision.
We wanted to find out how well the current dialog pipeline works in order to see where we should focus our further optimizations. Such a dialog requires a chain of individual operations, so the weakest member quickly dominates the overall quality. The most important elements of the chain are: Recording the speech via microphones, converting the speech into text, recognizing the intention, selecting a suitable response, converting the response into sound, output via loudspeakers. And then, of course, good timing is essential – you don’t notice it until it’s not there.
Some things worked very well, others not yet optimally. Even though there were occasional hang-ups, we were nevertheless very pleased that Navel was able to conduct relatively long and personal dialogs with several test subjects. In the film, we show an uncut example of 5 minutes of dialogue followed by a short interview with the test person about her assessment of Navel.