eduzhai > Applied Sciences > Engineering >

Speaker discrimination in humans and machines Effects of speaking style variability

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Does speaking style variation affect humans ability to distinguishindividuals from their voices? How do humans compare with automatic systemsdesigned to discriminate between voices? In this paper, we attempt to answerthese questions by comparing human and machine speaker discriminationperformance for read speech versus casual conversations. Thirty listeners wereasked to perform a same versus different speaker task. Their performance wascompared to a state-of-the-art x-vector PLDA-based automatic speakerverification system. Results showed that both humans and machines performedbetter with style-matched stimuli, and human performance was better whenlisteners were native speakers of American English. Native listeners performedbetter than machines in the style-matched conditions (EERs of 6.96 versus14.35 for read speech, and 15.12 versus 19.87 , for conversations), but forstyle-mismatched conditions, there was no significant difference between nativelisteners and machines. In all conditions, fusing human responses with machineresults showed improvements compared to each alone, suggesting that humans andmachines have different approaches to speaker discrimination tasks. Differencesin the approaches were further confirmed by examining results for individualspeakers which showed that the perception of distinct and confused speakersdiffered between human listeners and machines.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...