Answers are grounded in this paper and cite their sources.
Voice control is already firmly established in the consumer sector, but its potential for flexibility and efficiency is also increasingly being discussed in industry. For a secure application, e.g. to prevent exploitation by unauthorized machine operators, user verification based on automatic speaker recognition can be used as an additional authentication measure. The aim of this work is the evaluation of a speaker recognition system regarding various influences relevant for use in production scenarios. For this purpose, a data set of typical German production voice commands was used, which was recorded with four microphone configurations at ambient noise conditions up to 84 dB(A). Error rates regarding the acceptance or rejection of a speaker served as an evaluation measure for discussing optimal conditions for use in voice-controlled production. The greatest influence was seen in the choice of microphone configuration used, while channel mismatch and Lombard effect had a lesser influence on recognition. Furthermore, large improvements in error rates were observed up to an audio length of approximately 5 s. Under optimal conditions, over 90% correct acceptances and rejections at 84 dB(A) ambient noise level could be achieved. Further research is needed on the choice of an optimal decision threshold for accepting or rejecting speakers.
No takes yet. Share an insight, caveat, or question.
Norda et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: