TY - JOUR AU - Niesler, Thomas R. AU - Kamper, Herman PY - 2014/02/03 Y2 - 2024/03/28 TI - The impact of accent identification errors on speech recognition of South African English JF - South African Journal of Science JA - S. Afr. J. Sci. VL - 110 IS - 1/2 SE - Research Article DO - 10.1590/sajs.2014/20120049 UR - https://sajs.co.za/article/view/3643 SP - 1-6 AB - <div style="left: 288.189px; top: 208.571px; font-size: 15px; font-family: sans-serif; transform: scaleX(0.878665);">For successful deployment, a South African English speech recognition system must be capable of</div><div style="left: 288.189px; top: 228.586px; font-size: 15px; font-family: sans-serif; transform: scaleX(0.887787);">processing the prevalent accents in this variety of English. Previous work dealing with the different accents</div><div style="left: 288.189px; top: 248.601px; font-size: 15px; font-family: sans-serif; transform: scaleX(0.873122);">of South African English has considered the case in which the accent of the input speech is known. Here we</div><div style="left: 288.189px; top: 268.617px; font-size: 15px; font-family: sans-serif; transform: scaleX(0.880754);">focus on the practical scenario in which the accent of the input speech is unknown and accent identification</div><div style="left: 288.189px; top: 288.632px; font-size: 15px; font-family: sans-serif; transform: scaleX(0.842661);">must occur at recognition time. By means of a set of contrastive experiments, we determine the effect</div><div style="left: 288.189px; top: 308.647px; font-size: 15px; font-family: sans-serif; transform: scaleX(0.902797);">which errors in the identification of the accent have on speech recognition performance. We focus on the</div><div style="left: 288.189px; top: 328.662px; font-size: 15px; font-family: sans-serif; transform: scaleX(0.859701);">specific configuration in which a set of accent-specific speech recognisers operate in parallel, thereby</div><div style="left: 288.189px; top: 348.677px; font-size: 15px; font-family: sans-serif; transform: scaleX(0.878155);">delivering both a recognition hypothesis as well as an identified accent in a single step. We find that, despite</div><div style="left: 288.189px; top: 368.692px; font-size: 15px; font-family: sans-serif; transform: scaleX(0.862332);">their considerable number, the accent identification errors do not lead to degraded speech recognition</div><div style="left: 288.189px; top: 388.708px; font-size: 15px; font-family: sans-serif; transform: scaleX(0.897189);">performance. We conclude that, for our South African English data, there is no benefit of including a more</div><div style="left: 288.189px; top: 408.723px; font-size: 15px; font-family: sans-serif; transform: scaleX(0.883921);">complex explicit accent identification component in the overall speech recognition system.</div> ER -