ElevenLabs Text to Speech Cantonese: Voices, Cost & How to Use

ElevenLabs text to speech Cantonese runs on the v3 and v4 models — Cantonese (yue) sits in their supported-language lists (checked 5 Nov 2026) — but there is no dedicated Cantonese page, and the older Multilingual v2 and Flash v2.5 models exclude it. Generation costs 1 credit per character; the Free plan includes 10,000 credits per month.

There is no Cantonese page (checked 5 Nov 2026)

Let’s start with what I actually found, because most guides skip it. The URL https://elevenlabs.io/text-to-speech/cantonese returns a 404. The official languages sitemap (sitemap/languages__en.xml) contains no cantonese entry. And the language index on the text-to-speech hub page lists 90+ languages without Cantonese among them. All three checks were done on 5 Nov 2026. ElevenLabs supports Cantonese through its models, not through a dedicated marketing page — which changes how you verify claims about it.

This matters because language pages are where ElevenLabs publishes voice rosters, demo boxes, and use cases. With no Cantonese page, there is no official voice list, no sample demo text, and no published use cases for Cantonese specifically. Everything verifiable about Cantonese TTS comes from the model documentation instead — which is exactly where I went.

Which models actually handle Cantonese

The official model reference (elevenlabs.io/docs/overview/models, checked 5 Nov 2026) is explicit. Eleven v3 supports 70+ languages “including: … Cantonese (yue)…” — it names the three-letter code directly. Eleven v4 supports 90+ languages and its list also names “Cantonese (yue).” So the two current flagship models cover Cantonese. That’s the good news.

The older models do not. Multilingual v2 covers exactly 29 languages — en, ja, zh, de, hi, fr, ko, pt, it, es, id, nl, tr, fil, pl, sv, bg, ro, ar, cs, el, fi, hr, ms, sk, da, ta, uk, ru — and there is no yue in that list. Flash and Turbo v2.5 cover 32 languages, which is the same 29 plus Hungarian, Norwegian, and Vietnamese; again, no yue. Our Multilingual V2’s 29 languages are documented in our model guide if you want the full roster. Bottom line: for Cantonese you must select Eleven v3 or v4 in the app — v2 and Flash v2.5 will not do it.

Character limits differ by model: v3 allows 5,000 characters per request, v4 allows 10,000 (about 10 minutes of audio). Both support audio tags — the v3 docs describe them as controls for emotion, timing, and delivery, and v4 adds multi-speaker dialogue. If you’re comparing coverage across languages, our languages v4 actually covers page tracks the full list.

What Cantonese generation looks like without a language page

In practice you generate Cantonese the same way as any supported language: open Text to Speech in the ElevenLabs app, pick Eleven v3 or v4 as the model, paste your Cantonese text, choose a voice, and generate. There is no Cantonese preset or language picker entry to confirm against — the model selection is what determines language support. Download as MP3, or route longer builds through ElevenLabs Studio.

Cantonese is tonal, so small pitch errors change meaning — native listeners catch them instantly. Because there’s no official demo box to audition against, your testing burden is higher here than for languages with dedicated pages. Generate short samples first, listen for tone slips, and reword lines that the model keeps misreading. Each attempt costs credits, so short tests beat long regenerations.

ElevenLabs publishes no script-input guidance for Cantonese, so test your exact script in the app rather than trusting a guide (including this one) on traditional-character handling or written-Cantonese specifics.

Voices: the honest unknown

Here is the part other guides won’t say plainly: no official Cantonese voice roster exists. The language pages for Romanian, Japanese, Punjabi, and Indonesian each show their voice lineups — Cantonese has no page, so there is nothing to cite. Any article naming specific “Cantonese AI voices” on ElevenLabs is going beyond what ElevenLabs publishes.

What the docs do establish is that Cantonese (yue) is a supported v3/v4 language, so Cantonese text can be generated with voices from the library on those models. Keep the same voice across a series for consistency, and preview with short Cantonese sentences before committing to long generations. For dialogue-heavy content, the YouTube Shorts voiceover walkthrough shows how voice choice affects short-form output.

Cantonese credit math

Cantonese bills at 1 credit per character on v3 and v4 — the language never changes the rate. A 1,000-character script costs about 1,000 credits. Prices are in USD, checked 5 Nov 2026:

*First-month offers were shown until Nov 8 when checked on 5 Nov 2026. The Free plan’s 10,000 monthly credits cover about 10,000 Cantonese characters — enough for serious voice testing before paying. Credits reset monthly and don’t roll over. See our ElevenLabs pricing guide for creators for the complete tier comparison.

Pros and cons

Who should NOT use ElevenLabs for Cantonese

Don’t use it if you’re on Multilingual v2 or Flash v2.5 and can’t switch models — Cantonese simply isn’t in those models’ language lists, and no setting changes that. Don’t use it for certified broadcast dubbing, where a human voice actor’s performance and sign-off are the actual requirements. Don’t use it offline, since all generation runs in ElevenLabs’ cloud. And don’t build on secondhand claims about specific Cantonese voices — with no official roster, you’re testing blind, so budget credits for auditions. For Hong Kong YouTubers, advertisers, e-learning creators, and teams doing Cantonese localization drafts, the free tier is the honest way to find out if v3/v4 output meets your bar.

When the model keeps misreading a term, the pronunciation dictionary setup explains alias and phoneme rules that pin readings down without repeated regeneration.

Frequently asked questions

Does ElevenLabs support Cantonese text to speech?

Yes. Cantonese (yue) appears in the supported-language lists for Eleven v3 (70+ languages) and Eleven v4 (90+ languages) in the official model documentation (checked 5 Nov 2026). There is no dedicated Cantonese page on elevenlabs.io — support comes through the models, not a marketing page.

Why is there no Cantonese page on elevenlabs.io?

The /text-to-speech/cantonese URL returns a 404, the languages sitemap has no cantonese entry, and the TTS hub’s language list doesn’t show Cantonese — all verified 5 Nov 2026. ElevenLabs documents Cantonese in its model reference instead. That means no official voice roster, demo box, or use cases exist for Cantonese specifically.

Which model should I use for Cantonese?

Eleven v3 or v4 — they are the only current models whose supported-language lists include Cantonese (yue). Multilingual v2’s 29 languages and Flash/Turbo v2.5’s 32 languages both exclude it, so those models cannot generate Cantonese. Choose v4 for 10,000-character requests or v3 for 5,000-character requests with audio-tag direction.

Which voices read Cantonese?

No official Cantonese voice roster is published — with no language page, there is no lineup to cite. Cantonese text can be generated with voices from the library on v3/v4. Audition candidates with short Cantonese sentences first; tone errors are obvious to native listeners.

How much does Cantonese text to speech cost?

One credit per character on v3 and v4. The Free plan costs $0 with 10,000 credits per month (checked 5 Nov 2026); Starter is $6 per month with 30,000 credits after a $1 first month; Creator is $22 per month with 121,000 credits after an $11 first month; Pro is $99 per month with 600,000 credits.

Can I use Cantonese audio commercially?

Paid plans include commercial rights for generated audio — Starter ($6/month, checked 5 Nov 2026) lists a Commercial License, with broader rights on higher tiers. The Free plan is for non-commercial testing. Confirm the current license terms at checkout before client work.

What couldn’t you verify about Cantonese TTS?

Four things. Named native voices (no roster exists). Regional dialects such as Hong Kong versus Guangzhou speech (undocumented). Script-input behavior for traditional characters or written Cantonese (no published guidance). And tone-accuracy specifics (no official benchmarks). Everything else in this guide comes from the pages listed below.

Sources checked

Official pages checked on 5 Nov 2026:

• https://elevenlabs.io/text-to-speech/cantonese — returns 404 (no dedicated Cantonese page)
• https://elevenlabs.io/sitemap/languages__en.xml — no cantonese entry in the language sitemap
• https://elevenlabs.io/text-to-speech — language index shows no Cantonese among the 90+ languages
• https://elevenlabs.io/docs/overview/models — v3 (70+) and v4 (90+) include Cantonese (yue); v2’s 29 and Flash v2.5’s 32 exclude it
• https://elevenlabs.io/pricing — plan prices, credit allotments, and first-month offers