How is Canto-Gen different from big-name TTS?

Most big-name TTS adapts Mandarin models to Cantonese, resulting in flat tones, missing entering tones, and incorrect particles. Canto-Gen is trained from scratch for Cantonese, with accurate nine tones, code-mixing, and natural Hong Kong speech.

What's the difference between subscription characters and character packs?

Subscription characters reset monthly (e.g. Pro 600K chars/month)—unused quota expires. Character packs are one-time top-ups that never expire; use them at your own pace.

Will I lose credits if generation fails?

No. Failed generations are auto-refunded. Punctuation isn't counted—only actual generated characters are billed.

Can I use generated audio commercially?

Yes. Audio rights belong to you. Use them in videos, ads, audiobooks, products, and other commercial projects.

Is there API / MCP access?

Yes. Free tier includes Sandbox (20 calls/day). Paid plans have unlimited access. One-click prompt copy for Claude, Cursor, Copilot, and other agents.

港嘢 Canto-Gen

Canto-Gen is a Cantonese text-to-speech service built from the ground up for Hong Kong Cantonese. Trained on 8,000 hours of real conversations, it handles polyphones, particles, and code-mixing naturally. Offers voice cloning, SRT alignment, and API access for video dubbing, audiobooks, and smart home applications.

Visit 港嘢 Canto-Gen

Summary

Canto-Gen is a Cantonese text-to-speech (TTS) service designed specifically for Hong Kong Cantonese， solving the problem of generic Chinese TTS engines that produce flat tones， missing entering tones， and incorrect particles. Built by a local team， trained on 8,000 hours of real dialogue and 1 million slang/code-mixed phrases.

What is Canto-Gen?

Canto-Gen is a Cantonese voice generation system powered by the proprietary Canto-x1 model， engineered from scratch for Cantonese. It doesn't adapt Mandarin or generic Chinese models—it captures how Hong Kong people actually speak: pronunciation， intonation， slang， rhythm， and code-mixing. Nine tones and six pitches preserved; entering tones， liaison， and authentic Cantonese handled naturally.

Core Capabilities

Nine-tone accuracy: Polyphones ("助長" zoeng2 vs "長度" coeng4)， particles ("喇、囉、啦")， entering tone handling
Seamless code-mixing: "我哋下個 meeting 再傾" flows naturally without breaking at English words
Voice cloning: 15-second recording clones a specified voice， preserving Cantonese intonation
50+ voice library: Select by accent， age， style (young female， mature male， elderly， narrator， etc.)
SRT subtitle alignment: Direct SRT export with frame-accurate sync
API + MCP integration: One-click prompt copy for Claude， Cursor， Copilot， and other agents
20+ audio processing techniques: Rhythm， articulation， and intonation match natural Hong Kong speech

Pros

Authentic Cantonese: Sounds human， not robotic—"唔該晒"， "係咪"， "得喇" spoken naturally
Flawless code-mixing: Hong Kong colloquialisms like "個 draft 我擺左喺 share drive" pronounced correctly
Consistent voice: From narration to character dialogue， tone and rhythm stay on-brand
Transparent pricing: Pay per character (HK$0.49–0.64/1K chars)， usage-based， auto-refund on failed generation
Instant start: Free tier 5K chars/month (~6 min audio)， no credit card required

Positioning & best fit

Canto-Gen is built for creators， businesses， and developers who need high-quality Cantonese voice. It's not a generic TTS—it's a specialized tool deeply optimized for Hong Kong accent and context.

Content creators: YouTubers， podcasters， audiobook producers needing authentic Cantonese narration or character voices
Video production teams: Ads， shorts， tutorials requiring voice-over with SRT subtitle sync
Product developers: Smart home， chatbots， voice assistants needing API access and natural Cantonese output
Enterprise applications: Customer service systems， voice notifications， training materials requiring high-volume quotas and priority support