Cantonese data for AI training, collected to your brief
Cantonese data from IndecalAI is collected or reviewed by a native Cantonese speaker, not pulled from an existing archive. We work with speakers across East and Southeast Asia, where writing systems and speech patterns vary widely from one market to the next, matched to your brief and checked before delivery.
What Cantonese data can we produce?
Cantonese data from IndecalAI is speech, text, conversational and translation data collected or reviewed by a Cantonese speaker for your brief, not pulled from an existing archive.
Writing systems, formality levels and spoken rhythm shift noticeably across East and Southeast Asia, so a brief written for one market rarely transfers cleanly to the next without a native reviewer checking it.
This sits across several of our services depending on what you need: data collection for raw speech or text, conversational AI data for dialogue, and translation and localization for converting between Cantonese and another language.
Image and video categories on our datasets hub don't depend on language at all, so a project can combine this with visual data from the same region.
Cantonese data at a glance
- Data types
- Speech, text, conversational, translation
- Services
- Data collection, Translation
- Who works on it
- Contributors from our network of 30,000+ people worldwide, matched to your brief.
- Review
- A review pass before anything is delivered.
- Your data
- NDA first, used only for the agreed scope, returned or deleted when you ask.
- First step
- A small free sample before you commit.
What to know about Cantonese before you brief it
Cantonese is the everyday language of Hong Kong and Macau and of much of Guangdong, and it is the home language for many overseas Chinese communities in places such as Vancouver, London and San Francisco.
Cantonese is mostly spoken, and formal writing in Hong Kong usually follows standard written Chinese, close to Mandarin in grammar. When people type informally they use extra characters that exist only for Cantonese words. Decide whether transcripts should capture speech as it was said or convert it to the standard written form, because the two produce very different text. Tones carry meaning, so reviewers must be native listeners.
Languages that sit close to Cantonese in script, family or region: Mandarin Chinese data, Wu Chinese data, Vietnamese data, Thai data, Japanese data and Korean data.
Deciding what to specify for any language? The guide to speech data collection covers the choices that apply across languages, and the one on transcription for AI and for people covers how speech becomes text.
Cantonese at a glance
- Language family
- Sinitic (Chinese)
- Script
- Traditional Chinese characters, plus some Cantonese-only characters
- Spoken mainly in
- Guangdong and Guangxi in China, Hong Kong, Macau and overseas Chinese communities
- Name it in a brief
- Hong Kong Cantonese, Guangzhou Cantonese, overseas Cantonese
- Its own name
- 廣東話
Four ways to request Cantonese data
Each links to the service that runs it, with its full scope and process.
Cantonese speech data
Recorded speech in Cantonese: calls, prompts and voice bot utterances, matched to the region and reviewed before delivery.
Cantonese text data
Written Cantonese: messages, reviews, tickets or chat logs, written or labeled by people who read it natively.
Cantonese conversational data
Scripted or spontaneous Cantonese dialogue for voice bots, support agents and customer-facing assistants across this region's markets.
Cantonese translation
Human-verified translation between Cantonese and another language, covering prompts, UI copy and training data.
What to put in a Cantonese brief
Four things worth deciding before contributors start, so the first sample already looks right.
Region or dialect
Cantonese is spoken mainly in Guangdong and Guangxi in China, Hong Kong, Macau and overseas Chinese communities. Worth naming in a brief: Hong Kong Cantonese, Guangzhou Cantonese, overseas Cantonese. Tell us the region or community you're targeting so contributors are matched to the right one.
Script or transliteration
Script for Cantonese: Traditional Chinese characters, plus some Cantonese-only characters. Say whether you want the native script, a transliterated version, or both, especially if your downstream system only reads one.
Data type mix
Speech, text, conversational and translation can be requested together or on their own. Tell us the mix your model needs.
Volume and timeline
A rough sense of how much data and how soon lets us staff the right number of contributors from the start.
From your brief to reviewed Cantonese data
One named contact owns the project from scoping to the last file.
Brief
You tell us the data type, the Cantonese variety or region and roughly how much you need.
NDA and scope
We sign an NDA before any data moves, then agree the scope with you.
Free sample
We produce a small sample free, so you can check it before you commit.
Production
Contributors from our network work in our own tools, to the brief you set.
Review
A review pass runs before anything is delivered.
Delivery
The data comes to you, or moves on to another service with us.
Where Cantonese data ends up
Common jobs for Cantonese data. They're examples of the work, not a client list.
Voice bots and IVR
Recorded Cantonese prompts and calls teach a voice bot to recognize how people actually speak, not just how a script reads.
Support chat and tickets
Everyday Cantonese messages, written in the domain you work in, give a chat assistant real phrasing to learn from.
Content moderation and sentiment
Labeled Cantonese posts, reviews and messages teach a model to read tone and intent correctly in this language.
Localizing a product
Translating prompts, UI copy and training data into Cantonese so a product reads like it was written for that market, not machine-translated.
Off-the-shelf Cantonese data, or built to your brief?
We hold around 35,000 hours of off-the-shelf speech, prompts and conversations already collected across our network, and roughly 500 hours of it is Cantonese: see the Cantonese speech dataset. Here's how the two paths compare.
| Topic | A typical off-the-shelf dataset | Cantonese data with us |
|---|---|---|
| Already collected? | No way to know until you've already paid for it | Yes: roughly 500 hours of Cantonese speech are already recorded, and we tell you what they cover before you buy |
| Fit to your brief | Built for whoever bought it first, not your exact region, register or use case | Collected or reviewed to the region, script and format you specify |
| Who else has it | The same files can be sold to more than one buyer | Custom collections are scoped to your engagement. For stock data, who else can license it is set in your licence |
| Checking the fit | You find out after you've already paid for it | A small free sample first, so you check it before committing to volume |
Cantonese data terms, in plain words
General definitions. They describe the field, not a claim about us.
Where Cantonese data fits
Cantonese is one of 100 languages on our languages directory. You can also browse every category on the datasets hub.
Questions about Cantonese data
The datasets FAQ covers questions that apply across every category.
Do you support Cantonese?
Yes. Cantonese is one of 100 languages in this directory, part of our wider language coverage worldwide, including deep Indian language coverage. Tell us the regional variety you need.
Do you already have any Cantonese data collected?
Yes. Roughly 500 hours of Cantonese speech, prompts and conversations are already recorded and ready to license. The Cantonese speech dataset page has the details, and anything it doesn't cover we collect to your brief.
Can you mix regional varieties of Cantonese in one project?
Yes. Tell us the mix of regions or dialects you need and we'll match contributors from our network to that brief.
Which data types are available in Cantonese?
Speech, text, conversational data and translation to or from Cantonese. Image and video categories don't depend on language, so Cantonese speakers can also appear in those briefs on request.
Is a sample available before we commit?
Yes, a small free sample first.
Can this run alongside other languages in the same project?
Yes. Many engagements mix languages, domains and data types under one brief and one named contact, using our network of 30,000+ people across 150+ projects delivered so far.
What counts as a regional variety of Cantonese?
A version of the language tied to a place or community, which can differ in vocabulary, pronunciation, formality or script from another region's variety. Name the region in your brief if it matters for your use case.
Can you handle Cantonese in its native script?
Yes. Contributors work in the native script by default, with transliteration available if your project needs it instead.
Which variety of Cantonese should I ask for?
Ask for the one your users actually speak. Cantonese is spoken mainly in Guangdong and Guangxi in China, Hong Kong, Macau and overseas Chinese communities, and the forms most worth naming in a brief are Hong Kong Cantonese, Guangzhou Cantonese, overseas Cantonese. If you are unsure, describe your users and where they live, and your named contact will help you decide before collection starts.
How is Cantonese written, and does it matter for the data?
Cantonese data uses Traditional Chinese characters, plus some Cantonese-only characters. The script and its encoding affect how a model reads the data, so say in the brief which one you need and whether a romanized or transliterated version is also required.
Request Cantonese data
Pick the data type, confirm Cantonese as the language and a rough volume, and send it. We'll produce a small sample free before you commit.
- Your named contact replies within one business day.
- We sign an NDA before any data changes hands.
- We produce your free sample, then agree the full scope.
This opens your email app with the brief filled in. Nothing is sent until you press send. You can also use the contact page.