Text to speech · 廣東話 / 粵語
Pāṇini TTS takes Cantonese written the way Hong Kong people type it, in Traditional characters with colloquial forms like 係, 唔 and 嘅; there is no native Cantonese voice yet, so the samples are read by Xiǎoyàn, a Mandarin voice.
Sonex 速遞同你確認：你張 order 聽日下晝兩點至四點送到，記得屋企要有人收貨呀。如果時間唔啱，撳 SMS 入面條 link 就可以改期㗎喇。
Voices
Free to play in your browser without an account, up to 220 characters at a time. To save the audio as a file, sign up free for 10,000 characters and one voice clone, no card required: the dashboard saves WAV and the API returns MP3.
Making a video? See AI voice-over for YouTube and Reels. The audio is yours to use on monetised channels.
Last verified 24 October 2026
There is no native Cantonese platform voice yet, so the playground uses Xiǎoyàn, a voice recorded in another language: the Cantonese text is read correctly, in that voice's accent. For a local accent, clone a native Cantonese speaker from a 10-second recording and use that voice for every clip.
Sign up free, then send Cantonese text to the API with "output_format": "mp3" and save the file, or generate clips in the dashboard and download them. Scripts longer than the playground's 220 characters work the same way.
About Cantonese
Cantonese is usually described as having six tones (nine if the clipped checked syllables are counted separately), against Mandarin's four, and its vocabulary and grammar differ enough that the two are not mutually intelligible in speech. Formal writing in Hong Kong follows standard written Chinese, which reads like Mandarin; but messages, ads and chats are written in colloquial Cantonese with its own characters: 係 (is), 唔 (not), 嘅 (possessive), 咗 (completed action), 冇 (don't have), 佢 (he/she). Text meant to be heard on a call should be written in that colloquial form, or it sounds like a news bulletin.
Hong Kong Cantonese mixes English constantly, often mid-sentence: "book 咗 table", "check 吓 status", "send 個 email 畀你". Prices, flight numbers and English names stay in English. Hong Kong's banks, insurers, airlines and property firms run heavy call volumes, and many customers prefer Cantonese to Mandarin or English. Xiǎoyàn, a Mandarin platform voice, reads these samples; for a native Hong Kong accent, clone a Cantonese speaker from a 10-second sample.
Use cases
Banks and insurers confirm appointments and read payment reminders in the colloquial Cantonese customers speak.
Carriers and travel agencies handle booking changes and flight notices for Hong Kong and Macau travellers.
Clinics and services in Canada, Australia, the UK and Malaysia reach older Cantonese-speaking customers.
Commercial use
The full wording is in our terms of service.
YouTube
Yes, AI voices are allowed. What YouTube limits is mass-produced, repetitive content: videos made from a template with little original work. Its July 2025 update renamed that rule from “repetitious” to “inauthentic” content (YouTube’s monetisation policies).
Write your own script, add your own research or editing, and keep one consistent voice for the channel. Where a realistic AI voice could mislead viewers, use YouTube’s altered or synthetic content label. The monetisation decision is always YouTube’s.
Voice cloning
Clone your voice from a 10-second recording. Your clone speaks Cantonese, and every other language Pāṇini supports, in your own voice. The language comes from the text you type, so a clone recorded in English or Hindi can read a Cantonese script.
Your accent comes with it: a clone sounds like you speaking Cantonese, not like a different native speaker. Every free account includes one voice clone. Clone only your own voice, or a voice whose owner has agreed.
API
One request returns Cantonese audio as MP3. The language is detected from the text, so no language setting is needed. Pāṇini TTS starts streaming audio in under 200 ms (time to first byte).
1curl https://api.sonexlabs.com/v1/speech \2 --header 'Authorization: Bearer YOUR_API_KEY' \3 --header 'Content-Type: application/json' \4 --data '{"text": "Sonex 速遞同你確認：你張 order 聽日下晝兩點至四點送到，記得屋企要有人收貨呀。如果時間唔啱，撳 SMS 入面條 link 就可以改期㗎喇。", "output_format": "mp3"}' \5 --output cantonese.mp3FAQ
Yes. The playground on this page reads up to 220 characters of Cantonese at a time with no account. Sign up free for 10,000 characters and one voice clone, no card; after that it costs $0.03125 per 1,000 characters.
Yes, after free sign-up. The playground on this page plays audio in your browser only, up to 220 characters at a time, so you can check the voice first. Sign up free (10,000 characters and one voice clone, no card): the dashboard saves each clip as a WAV file with Download audio, and the API returns MP3 when you set "output_format": "mp3".
Yes. Cantonese has its own tones, words and grammar, so write colloquial Cantonese for a call. The platform voice on this page is a Mandarin voice; for a Hong Kong accent, clone a native Cantonese speaker from a 10-second sample.
Yes, along with Traditional and Simplified characters and English words mixed into the sentence.
Xiǎoyàn reads these samples; any Sonex Labs voice can speak Cantonese, and you can clone a native speaker's voice from 10 seconds.
$0.03125 per 1,000 characters, the same as every language.
Yes. Audio you generate, on the free tier or with paid credits, can be used on monetised YouTube channels, Shorts, Reels and in ads, with no attribution. YouTube limits mass-produced, repetitive videos, not AI voices, so write your own script and keep your edits original. The monetisation decision is YouTube's.
The audio you generate is yours to use, including commercially, with no attribution and no royalties to Sonex Labs. You are responsible for the script itself: if you read someone else's text, you need their permission.
Yes. A Sonex voice clone is not tied to the language it was recorded in. Type a Cantonese script and your clone reads it in your own voice, with Cantonese pronunciation. Your natural accent carries over, so it sounds like you speaking Cantonese.
More languages
See text to speech in every language, the full list of 250+ languages, pricing, the Pāṇini TTS API, or transcribe Cantonese audio with Valmiki speech to text.
10,000 free text-to-speech characters and one free voice clone that speaks every language, no card required. Every language is billed at the same rate.