0
HHassan Kamal提问 · Jul 10, 2026, 09:56 AM
Voice NOC in Arabic is fantastic for our Gulf staff but our Egyptian engineers get maybe 60% intent recognition on the same phrases. Docs mention regional model tuning is coming in 1.3 but has anyone found a workaround in 1.2? A per-user speech profile setting perhaps? We've considered training our team to use MSA but that's a hack.
#voice-noc#arabic#ai-engine#regional
2 条回复·533 次浏览
0
SSamir Al-Otaibi回复 · Jul 12, 2026, 11:55 AM
Egyptian dialect handling is genuinely rough in 1.2. Two things that helped us: (1) turn on the 'preferred phrases' setting per user so common commands map to intent regardless of exact wording; (2) ask engineers to speak short, verb-first commands (e.g. 'reset ACS Karachi' vs conversational 'can you reset the ACS in Karachi'). Not glamorous but tides you over until 1.3.
0
NNabil Zaidi回复 · Jul 13, 2026, 09:28 AM
Confirmed from a Netxol side — Egyptian and Levantine dialect models are in 1.3 beta already. Sign up at /downloads for the Cirrus beta if you can run it on a staging Core, we're actively collecting recognition-failure samples.