Transcribe Cantonese Audio to Text (2026)

Cantonese is spoken by more than 85 million people, and on iPhone it barely has a transcription path at all. Apple's Voice Memos doesn't support Cantonese — Mandarin is the only Chinese dialect in its ten-language list. WeChat's 转文字 button is trained on Mandarin. iOS keyboard dictation only produces Cantonese text if the keyboard language is explicitly set to Cantonese, a setting almost nobody enables. If you want Cantonese audio turned into Chinese characters, the built-in options are effectively nonexistent.

Audio Notes treats 廣東話 as a first-class language. It sits in the picker's popular row — English, 中文, 廣東話, 日本語, 한국어, Español — as the one language added on top of Whisper's 99, making 100 entries total. Cantonese is handled as its own language, not as a dialect of Chinese.

Now the honest part, because this page earns its keep by being straight: Whisper is the best open model for Cantonese, and even it is the weakest of the major languages at roughly 13% word error rate. The turbo model we run server-side (large-v3-turbo, with large-v3 as fallback) degrades a little more on Cantonese than full large-v3 does, and auto-detect is at its weakest for Cantonese — so set 廣東話 explicitly, keep clips short and clean, and expect draft-quality text that's excellent for notes and translation, not for verbatim publication.

For a deeper look at the platform-level gaps, our Cantonese voice-to-text guide in 繁體中文 covers the history of the problem in detail. This page focuses on what the app actually does.

Getting Cantonese Audio into Characters

1. Record or import

Open the app and pick 廣東話 from the language picker — it's in the popular row, not buried in a submenu, and it's the one entry added on top of Whisper's 99 languages (100 total). Record live for up to 90 minutes per clip, or import audio in m4a, mp3, wav, caf, aac, aiff or mp4a; video in mp4, mov, m4v, avi or mkv works too, with the audio track extracted automatically. Free accounts get 60 minutes of transcription in total — lifetime, not monthly; Pro is $5.99 a month for 10,000 minutes. Transcription runs server-side on Whisper, so internet is required and nothing is processed on the phone — the same pipeline that powers transcribing audio in any language across all 100 entries.

2. Share from WhatsApp

The share sheet is the lifeline for Cantonese families, especially the Hong Kong diaspora. A voice note from a relative arrives in most apps as a garbled best guess; here it becomes 廣東話 characters. Share the message into Audio Notes and it transcribes like any other clip — the piece most WhatsApp voice-to-text setups are missing is a model that actually handles Cantonese. WeChat voice notes follow the same share path.

3. Why the built-in options fail

Worth spelling out exactly what iOS can't do, because most Cantonese speakers assume the phone handles it:

Voice Memos transcription supports ten languages — English, Spanish, Portuguese, Italian, French, German, Japanese, Korean, and Simplified and Traditional Chinese — so transcribing Mandarin audio works natively, but Cantonese is not on the list at all.
WeChat's 转文字 button is built for Mandarin; on Cantonese audio it either fails or returns Mandarin text.
iOS keyboard dictation writes Cantonese only when the keyboard language is set to Cantonese — a hidden setting most people never touch — and even then the text can't be translated, exported, or searched.

4. Then: English translation

Because English is the app's preferred language, every Cantonese transcript is automatically translated into English and summarized in English. The Chinese characters stay underneath — translation is an extra layer, not a replacement. For a partner or colleague who doesn't speak Cantonese, that English layer is the whole point of the exercise.

Cantonese Transcription: The Options

Method	What you get	Limits	Best for
Record in Audio Notes	Live 廣東話 transcript in Chinese characters; English translation and summary	~13% WER — weakest of the major languages; set 廣東話 explicitly; 90-min cap; 60 free lifetime minutes	Clean, short clips: family notes, interviews, single answers
Import a file	Transcript of m4a, mp3, wav, caf, aac, aiff, mp4a audio and mp4, mov, m4v, avi, mkv video	iOS only; server-side, so internet required	Recorded interviews, TV clips, older audio files
Share sheet	WhatsApp, WeChat and other voice notes transcribed in place	Source app must expose Share; no speaker labels or diarization	Family WhatsApp messages, diaspora group chats
iOS built-in options	Effectively nothing: no Voice Memos Cantonese; Mandarin-biased WeChat 转文字	No reliable path to characters, translation, or reuse	Confirmation that a purpose-built app is needed

Cantonese Transcription: The Honest Answers

Does iOS transcribe Cantonese at all?

Effectively no. Voice Memos transcription covers ten languages and Cantonese isn't one of them. iOS keyboard dictation can write Cantonese characters only if the keyboard language is set to Cantonese — a hidden, rarely used setting — and even then the text can't be translated or exported. Audio Notes is one of the only first-class Cantonese transcription paths on iPhone, running Whisper server-side.

How accurate is Cantonese transcription?

Draft-quality, not publication-quality. Whisper is the best open model for Cantonese at roughly 13% word error rate — the weakest of the major languages, and we're equally honest about the other Southeast Asian entry in the picker in the transcribe Vietnamese audio guide. The turbo variant degrades a bit more on Cantonese than full large-v3, so pin 廣東話 and keep clips short and clean. For notes, translation, and summaries it's genuinely useful; for verbatim quotes, proofread.

Can I transcribe Cantonese to English?

Yes. The transcript comes out in Chinese characters, and because English is the app's preferred language it's auto-translated and summarized in English — the same flow the hub describes for transcribing audio in any language, applied to Cantonese. The characters stay underneath.

Can I transcribe WeChat or WhatsApp voice messages in Cantonese?

WhatsApp: yes — share the voice note into Audio Notes and it comes out as 廣東話 characters; the transcribe voice messages guide covers the exact share flow. WeChat: also yes via the same share path — but use Audio Notes rather than WeChat's own 转文字 button, which is Mandarin-first and will not give you Cantonese text.

Is the transcript in written Chinese or romanization?

Written Chinese characters — Whisper's Cantonese output is character-based, so the transcript is 中文 text, not Jyutping or Yale romanization. Romanization isn't offered as an output format; if you need it, translate the transcript to English or convert the characters yourself.

Audio Notes AI turns recordings in any language into text you can search, copy, translate, and summarize — on iPhone. Every account includes 60 minutes of lifetime transcription for free; Pro ($5.99/month) unlocks 10,000 minutes per month with a 90-minute cap per recording. Whisper-powered transcription runs server-side in 60+ languages, transcripts are translated and summarized automatically when the detected language differs from your preferred language, and shared notes are encrypted with AES-256 on your iPhone before upload — the decryption key lives only in the link.

More Language Guides

Download Audio Notes AI free on the App Store — 60 minutes of lifetime transcription included.