How does the NLU training pipeline actually handle Australian English phonetics when pushing to the Sydney edge? The deployment pipeline stalls on the intent mapping step for bot-au-ivr-09. It’s running v2024.3.1, and the conversational bot keeps failing validation. Training dataset holds about 2000 utterances, heavily weighted toward +61 mobile prefixes. Deployment throws a 422 Unprocessable Entity with an ACMA compliance flag attached. Logs show the Sydney edge timing out on phoneme extraction, probably hitting that 300ms latency ceiling.
Cleared the cache, re-exported the utterance CSV, and switched the language locale to en-AU. Nothing sticks. Flow validation passes locally, but the cloud deployment barfs. ACMA rules require explicit caller ID masking before NLU processing kicks in, yet don’t see any masking blocks in the architect flow yet. Just a raw training run.
- Org domain: mypurecloud.com.au
- Edge: Sydney (ap-southeast-2)
- Platform version: 2024.3.1
- Bot ID: bot-au-ivr-09
- NLU Engine: Genesys Conversations AI v2
- Steps attempted: Rebuilt intent corpus, adjusted phonetic normalization rules, toggled ACMA recording compliance flag in bot settings
POST /api/v2/conversations/ai/bots/bot-au-ivr-09/nlu/train returns
{
“code”: “bad_request”,
“message”: “Phoneme extraction failed for en-AU locale. ACMA compliance validation requires explicit caller ID masking before NLU processing.”,
“details”: “Latency threshold exceeded on sydney-edge-02. Timeout after 312ms.”
}