Australian-English clips the cloud STTs drop. Camlin gets them right.
Camlin Speech recognises “Centrelink CRN”, “rego renewal”, “NBN FTTP”, “BPay reference”, “Cairns”, “Mackay”, and the surname “Nguyen” more reliably than any cloud STT you can buy today — and runs on-prem on hardware you already own.
Where Camlin pulls ahead
“My Centrelink CRN is two-zero-two-six dash zero-zero-one-nine-eight.”
Generic STT typically transcribes "Centrelink" as "central link", or splits the CRN into prose. Camlin nails both because the AU vocab is in the training set.
“My rego is due next fortnight.”
"Rego" doesn't exist in standard US/UK English. Generic STT reaches for "rigor" or "ago". Camlin has it as a first-class word.
“My NBN FTTP is dropping out twice a day.”
"NBN" / "FTTP" / "FTTC" / "RSP" — Telstra, Optus, TPG and NBN Co care about these. Generic STT mangles them; Camlin gets them right the first time.
“My name is Aanya Reddy. R-E-D-D-Y.”
Open-vocabulary surnames are the structural problem no amount of training fixes alone. Camlin's per-tenant lexicon biasing solves it: register the surname once, recogniser primes itself for every call.
Want the side-by-side blind listening test with hypotheses from Camlin, Google, Azure, Deepgram and OpenAI Whisper? Open the live A/B (buyer-vocab tab).
Held-out AU buyer-vocab eval
Measured against a held-out buyer-vocabulary eval set covering 13 AU-specific categories — the phrases AU contact centres handle every day. Numbers refresh with every model release; the comparison page always shows the latest figures, no cherry-picking.
Source-clip provenance
Every clip on the live A/B page links back to its source audio, so you can verify what was played, who voiced it, and what the canonical transcript is. No cherry-picking.
Built and hosted in Sydney
The whole stack runs in ap-southeast-2 or on-prem on customer hardware. Audio never leaves your jurisdiction. The procurement story for Australian government and Tier-1 telcos.