topsystem, an EPG (Ehrhardt + Partner Group) company, asked itself this question and tested Lydia Voice for recognition reliability with a mask in a computer-aided test setup. "The result right away: when working with a voice-guided mask, the advantages of the speaker-independent LydiaVoice system are once again evident compared to systems based on a personal voice profile," says Tim Just, CEO of topsystem. "We have put a lot of development work into the current version of our voice software and this is paying off during the crisis: With our technology, we are helping logistics companies to maintain their operations and not have to accept any loss of productivity in order picking despite safety measures."

The current version of Lydia8, with which the test was carried out, is based on neural networks and works according to the deep learning concept. This has increased recognition reliability by 25%, which is a decisive advantage, especially for employees with a dialect or strong accent. "It is now clear that this precise recognition is possible even with a face mask without any adjustments to the system. In addition, no language training - and therefore no follow-up training with a mask - is necessary, as can be the case with speaker-dependent pick-by-voice solutions," adds Tim Just.

The test setup was standardized and computer-aided to determine the extent to which wearing an N95, FFP2, paper or fabric mask has an impact on the speech recognition rate with the current version of Lydia Voice. Tests were initially carried out with three different types of headset. The results were approximately the same for all variants with a mask compared to picking without a mask. Even when FFP2 masks are used, there are virtually no measurable losses with LydiaVoice. Picking with the Lydia VoiceWear also passes the mask test: in another test setup, results comparable to those with a headset were achieved. "Only when using FFP2 masks in combination with the Lydia VoiceWear are slight changes noticeable - but still within the tolerable range," adds Tim Just. "This deviation results from the fact that FFP2 masks have a greater influence on the acoustic signals due to their design."