Text to speech
Cantonese text to speech
Cantonese narration and voiceovers for Hong Kong and diaspora audiences. Giggy generates Cantonese speech for $5 per million characters on Fast and $10 on Streaming, the same price as every other language.
Text to speech: what you pay
| Provider | Mode or model | USD per 1M characters |
|---|---|---|
| Giggy | Fast | $5.00 |
| Giggy | Streaming | $10.00 |
| Fish Audio | S2.1 Pro | $15.00 |
| Inworld | Realtime TTS-2 Flash | $15.00 |
| OpenAI | tts-1 | $15.00 |
| Azure | Neural TTS | $15.00 |
| Inworld | Realtime TTS-2 | $25.00 |
| Amazon | Polly Generative | $30.00 |
| Deepgram | Aura-2 | $30.00 |
| Chirp 3: HD | $30.00 | |
| OpenAI | tts-1-hd | $30.00 |
| Cartesia | Sonic-3.6 · Scale overage | $38.00 |
| ElevenLabs | Flash v2.5 | $40.00 |
| ElevenLabs | Multilingual v2 / v3 / v4 | $80.00 |
| MiniMax | Speech-02-HD | $100.00 |
ElevenLabs, OpenAI, Deepgram, Google, Amazon and Azure checked November 11, 2026; Fish and Inworld October 20; Cartesia and MiniMax October 13.
Hear Cantonese voices
Cantonese intelligibility, published
CER when generated speech is transcribed back, on the same 100 prompts per voice. Lower is better.
| Model | CER |
|---|---|
| Giggy Streaming | 23.9% |
| MiniMax Speech-02-HD | 34.1% |
| ElevenLabs Multilingual v2 | 51.5% |
Generate Cantonese speech in three steps
Enter your Cantonese text
Type or paste your Cantonese script. Add expression tags such as [whisper] or [short pause] to guide the delivery.
Choose a Cantonese voice
Pick a Cantonese voice from the Voice Library or create your own, then choose Batch, Fast or Streaming.
Download or call the API
Download 24 kHz MP3 audio from the app, or request the same voice from the Speech API in your product.
Built for a wide range of uses
Voice agents
Give Cantonese assistants and support agents a natural voice with Streaming audio.
Podcasts and audiobooks
Narrate Cantonese chapters and episodes with a voice listeners can stay with.
Video voiceovers
Voice Cantonese explainers, ads and social videos without a recording booth.
E-learning
Turn Cantonese lessons and training into clear, consistent narration.
Games and characters
Give Cantonese characters distinct voices, with emotion and non-verbal tags.
Accessibility
Read Cantonese articles, documents and interfaces aloud.
One price in every language
Giggy counts Unicode code points in the full input text, including spaces and punctuation, at the same rate in all 240+ supported languages. Credits are prepaid with no subscription and shared across text to speech, speech to text and voice creation.
Shape the delivery
Add expression tags such as [whisper], [excited] or [short pause] to your Cantonese script. Giggy supports 50+ emotions, 20+ delivery styles and 40+ non-verbal events, plus speaking-speed control.
Build Cantonese speech into your product
Send Cantonese text to /v1/text-to-speech with a voice ID and your API key, or use the OpenAI-compatible speech endpoint or the Giggy MCP server. Batch and Fast return MP3; Streaming returns progressive 24 kHz PCM.
Frequently asked questions
What is Cantonese text to speech?
Text to speech turns written Cantonese text into natural-sounding audio. Giggy generates Cantonese speech in the app or through the API with voices made for Cantonese.
How much does Cantonese text to speech cost?
Giggy Fast costs $5 and Streaming main0 per 1,000,000 characters with Pro, Studio or Scale credits, the same as every other language. Batch is free within request and queue limits.
Where can I hear Cantonese voices?
Listen to the top Cantonese voices on this page or browse every Cantonese voice in the Voice Library, with samples, accents and suggested uses.
Can I create a custom Cantonese voice?
Yes. Voice Design creates a new Cantonese voice from a written description, and Voice Clone creates one from 3 to 10 seconds of authorized Cantonese speech. Each costs 1,000 credits.
Can I use Cantonese voices through an API?
Yes. Every library voice has a voice ID for the Giggy Speech API, its OpenAI-compatible speech endpoint and the Giggy MCP server.