Cantonese meeting transcription, written up for you.

Cantonese (粵語, colloquially 廣東話) is a distinct spoken language, not a Mandarin accent — and that distinction is exactly where most transcription tools fail. It carries six to nine tones depending on how you count them, a large stock of vocabulary and grammatical particles that never appear in Mandarin (係, 唔, 嘅, 咗, 啦, 喎, 㗎), and a colloquial spoken register that diverges sharply from the written standard. When someone speaks Cantonese in a Hong Kong boardroom, the words leaving their mouth are frequently not the words that would be written down in a formal document — and an engine trained to normalise everything toward written Standard Chinese will silently rewrite what was actually said.

The bigger trap is defaulting. Ask a general-purpose ASR system for 'Chinese' and it almost always routes audio through a Mandarin model, because Mandarin dominates training data and is the assumed prestige variety. Fed Cantonese, that model doesn't return an error — it returns confident, fluent, wrong Mandarin. Tones get flattened, particles get dropped or reinterpreted, and colloquial Cantonese phrases get 'corrected' into Mandarin equivalents that no one in the room said. You end up with a transcript that reads plausibly but misattributes meaning, which in a legal or financial setting is worse than no transcript at all. GetNotes selects Cantonese explicitly and keeps it there.

Then there's the code-switching, which in Hong Kong is not an edge case — it's the default speech pattern. A single sentence in a real HK meeting routinely runs Cantonese grammar with English nouns and verbs threaded through it: '你哋個 deadline 幾時 confirm?', 'follow up 完 send 個 proposal 俾我', 'ke case 有啲 risk 要 flag'. Legal, banking, and tech professionals switch to English for terms of art the way other people reach for jargon, because the English term is the term. A transcriber that treats each language as a separate mode breaks on every switch. GetNotes handles the mixing mid-sentence rather than forcing the audio into one language lane.

Increasingly there's a third language in the room too. Post-2000s Hong Kong meetings — and virtually every meeting on the Guangdong side of the border — mix Cantonese with Mandarin (Putonghua), especially when mainland clients, colleagues, or suppliers are present. Speakers slide between the two based on who's talking, and the two Chinese varieties share a writing system while sounding almost nothing alike. Distinguishing 'this speaker just switched from Cantonese to Mandarin' from 'this is still Cantonese' is precisely the judgement a Mandarin-default system can't make, because to it everything already is Mandarin.

Script matters for the output, not just the input. Hong Kong, Macau, and the Cantonese diaspora read and write Traditional Chinese, so GetNotes renders transcripts and AI minutes in Traditional characters (繁體字) rather than the Simplified set used on the mainland — a detail that quietly signals whether a document was produced for your audience or run through a mainland pipeline. Where a speaker uses genuinely colloquial Cantonese written forms, they're preserved rather than laundered into formal written Chinese, so the transcript reflects the meeting and the AI minutes present the clean, decision-focused summary.

The wedge under all of this is the room itself. Cantonese business still happens heavily face-to-face — the property viewing, the factory floor, the family dinner where the succession actually gets decided, the client sitting across your desk in Central. No meeting bot dials into any of those, because there's no call to dial into. GetNotes records from the phone already in your pocket through the browser — no app to install, no bot to invite, no awkward 'do you mind if I add an AI?' — captures the multi-speaker, multi-language conversation as it really sounds, and stores it privately with delete-anytime control. It owns the physical Cantonese-speaking room that a call-bot structurally cannot enter.

Where it earns its keep

How GetNotes works

1. Record or upload. Hit record on your phone or laptop in the room — or drop in an audio file you already have. No bot joins anything.
2. It writes it down. Every word transcribed, every speaker labelled, in 90+ languages including ones that switch mid-sentence.
3. Share the minutes.Summary, decisions, and action items with owners — ready before everyone's back at their desk.

Questions, answered

Other languages