# Best Chinese text to speech for apps, videos, and voice agents

Which Mandarin Chinese text to speech service fits an app, a narrated video, a voice agent, mixed Chinese and English text, or a brand voice, from each vendor's own docs.

Published 2026-11-17. Paxa Labs.

The best Chinese text to speech depends on the job. Paxa TTS Flash v2 reads Simplified Chinese, English, and Thai in one voice. MiniMax, ElevenLabs, Fish Audio, Cartesia, Azure AI Speech, Google Cloud, Amazon Polly, Murf, Narakeet, and WellSaid Labs also document Mandarin. MiniMax, Narakeet, and WellSaid list Cantonese as well.

Every vendor fact below links that vendor's own documentation, as our comparison pages verified it in October and November 2026. No vendor is scored or ranked. Published rates for each one are on [the text to speech API comparison](https://paxalabs.com/compare/text-to-speech-apis), re-checked every month.

## Which text to speech services speak Mandarin Chinese?

| Service | How its documentation lists Chinese | Also documented |
| --- | --- | --- |
| Paxa TTS Flash v2 | Mandarin from Simplified Chinese text, in all 50 voices | The same voices read English and Thai. Live WebSocket, and word, sentence, or utterance timestamps |
| MiniMax | Chinese and Cantonese as separate language values, among 40 | Voice cloning and realtime streaming |
| ElevenLabs | Mandarin Chinese among 90+ languages of its newest models | Library, cloned, and designed voices. HTTP and websocket access |
| Fish Audio | Chinese named among the 80+ languages of S2 Pro | Voice cloning and a streaming WebSocket |
| Cartesia | Chinese (zh) among 44 languages, with the zh-CN locale | A websocket endpoint and an add_timestamps parameter |
| Azure AI Speech | zh-CN voices in its Standard, Multilingual, and HD tiers | A WordBoundary event for each spoken word |
| Google Cloud | Mandarin (cmn-CN) in Preview on Gemini-TTS, and cmn-CN voices of the Chirp 3 HD, WaveNet, and Standard types | English and Thai as generally available on Gemini-TTS |
| Amazon Polly | One Chinese (Mandarin) voice, Zhiyu | Speech marks for words and sentences |
| Murf | A Mandarin (China) section with four Falcon 2 and six Gen2 voices | Falcon 2 streams for voice agents. ALAW and ULAW output |
| Narakeet | Separate voice lists for Mandarin Chinese, Taiwanese Mandarin, and Cantonese | SRT and VTT subtitle files with the audio |
| WellSaid Labs | Chinese (Cantonese and Mandarin), with six Mandarin (Simplified) partner voices | Word timing as JSON, SRT, and VTT |

Two widely used speech APIs leave Mandarin out. [Deepgram's Aura voices](https://developers.deepgram.com/docs/tts-models) cover seven languages, and Chinese is not one of them. [SpeechifyAI](https://docs.speechify.ai/build/guides/text-to-speech/language-support) lists six languages for simba-3.0 and English alone for simba-3.2. [OpenAI's guide](https://developers.openai.com/api/docs/guides/text-to-speech) says its speech models generally follow a 57-language list that includes Chinese, and describes its built-in voices as currently optimized for English.

## Which one fits an app or a website that speaks Mandarin?

Every Paxa TTS Flash v2 voice reads Mandarin from Simplified Chinese text, and 2 of the 50 are cast for text that is Chinese throughout. [The Chinese voices page](https://paxalabs.com/chinese-voices) plays them on one line. An app already written for the OpenAI SDK can call [the compatible endpoint](https://paxalabs.com/docs/openai-compatibility) by changing the base URL and key.

On the cloud platforms, [Microsoft's voice table](https://learn.microsoft.com/en-us/azure/ai-services/speech-service/language-support?tabs=tts) lists zh-CN voices in the Standard, Multilingual, and HD tiers of Azure AI Speech. [Google's voice list](https://docs.cloud.google.com/text-to-speech/docs/list-voices-and-types) has cmn-CN voices of the Chirp 3 HD, WaveNet, and Standard types, and [Gemini-TTS](https://docs.cloud.google.com/text-to-speech/docs/gemini-tts) lists Mandarin in Preview. [Amazon Polly's voice table](https://docs.aws.amazon.com/polly/latest/dg/available-voices.html) lists one Mandarin voice, Zhiyu, on its Neural and Standard engines.

[MiniMax's API](https://platform.minimax.io/docs/api-reference/speech-t2a-http) takes Chinese as one of 40 language values. [Botnoi Voice](https://voice.botnoi.ai/) lists Chinese among more than 20 languages and links a page of Chinese speakers.

Check how each vendor counts a Chinese character before you estimate a bill. [Microsoft's billing section](https://learn.microsoft.com/en-us/azure/ai-services/speech-service/text-to-speech#billable-characters) counts each Chinese character as two. Paxa TTS Flash v2 counts one character as one, at the same rate as English and Thai.

## Which one fits narrated videos, courses, and captions?

Captions need the time of each word as well as the audio. Paxa TTS Flash v2 returns [word, sentence, or utterance timestamps](https://paxalabs.com/docs/timestamps) on the same request as the audio, and [the subtitles page](https://paxalabs.com/subtitles) shows them on a caption line. One request reads up to 20,000 characters, and a longer lesson goes out as several requests whose audio you join.

[Narakeet's API](https://www.narakeet.com/docs/automating/text-to-speech-api/) can return SRT and VTT subtitle files with the audio, synchronized on paragraph boundaries. [WellSaid's word timing endpoint](https://docs.wellsaidlabs.com/reference/word-timing) returns the audio with word-level timing as JSON, SRT, and VTT files, and [its Global Voices page](https://docs.wellsaidlabs.com/reference/global-voices) lists six Mandarin (Simplified) partner voices. [Microsoft's Speech SDK](https://learn.microsoft.com/en-us/azure/ai-services/speech-service/how-to-speech-synthesis) raises a WordBoundary event at each spoken word. [Amazon Polly's speech marks](https://docs.aws.amazon.com/polly/latest/dg/speechmarks.html) mark where words and sentences start and end. [Cartesia's API](https://docs.cartesia.ai/api-reference/tts/tts) documents an `add_timestamps` parameter.

[What word-level timestamps are](https://paxalabs.com/guides/word-level-timestamps-api) explains what to ask a vendor for. A video that already has a voice track in another language is a dubbing job, and [the AI dubbing guide](https://paxalabs.com/guides/how-to-dub-a-video-with-ai) walks through one.

## Which one fits a voice agent or a phone line?

A voice agent speaks a reply while the language model is still writing it. The Paxa [live WebSocket](https://paxalabs.com/docs/live-streaming) takes text fragments as they arrive, and a cancel frame stops the reply when the caller talks over it. For phone lines, Paxa TTS Flash v2 returns G.711 μ-law and A-law at 8 kHz directly, beside PCM and WAV at 8 to 48 kHz.

[MiniMax's Speech-02 announcement](https://www.minimax.io/news/speech-02-series) describes realtime streaming and a Turbo variant for real-time use. [Cartesia's API](https://docs.cartesia.ai/api-reference/tts/tts) has a websocket endpoint, and [ElevenLabs' API reference](https://elevenlabs.io/docs/api-reference/introduction) offers websocket access. [Fish Audio's streaming guide](https://docs.fish.audio/features/realtime-streaming) describes a WebSocket that buffers text until it has enough context. [Murf](https://murf.ai/api/docs/introduction/overview) names Falcon 2 as its streaming model for real-time voice agents, with four Mandarin voices in [its voice library](https://murf.ai/api/docs/voices-styles/voice-library) and [ALAW and ULAW output at 8000 Hz](https://murf.ai/api/docs/text-to-speech/overview).

[The voice agent speech guide](https://paxalabs.com/guides/best-text-to-speech-api-for-voice-agents) compares live connections in more detail. [The realtime speech page](https://paxalabs.com/realtime-tts) and [the voice agents page](https://paxalabs.com/use-cases/voice-agents) cover the Paxa side.

## Which one reads Chinese mixed with English or Thai?

Product names, model numbers, and addresses put English inside a Chinese sentence. Every Paxa TTS Flash v2 voice reads Chinese, English, and Thai, and a phrase in one language inside a sentence in another goes through in the same request. A store serving Thai and Chinese customers keeps one voice across both, and [the Thai and Chinese page](https://paxalabs.com/thai-chinese) shows that pairing.

[ElevenLabs' models page](https://elevenlabs.io/docs/overview/models) describes Eleven v4 as keeping voice quality consistent across language switches. [Microsoft's voice table](https://learn.microsoft.com/en-us/azure/ai-services/speech-service/language-support?tabs=tts) describes multilingual voices that detect the input language. [MiniMax's API](https://platform.minimax.io/docs/api-reference/speech-t2a-http) sets the language as a value on the request.

Cantonese and Taiwanese Mandarin are separate needs. Paxa TTS Flash v2 speaks Mandarin and offers no Cantonese voice. MiniMax sets Cantonese as its own language value, [Narakeet](https://www.narakeet.com/languages/) keeps separate voice lists for Taiwanese Mandarin and Cantonese, and [WellSaid's API FAQ](https://docs.wellsaidlabs.com/docs/frequently-asked-questions) lists Cantonese beside Mandarin. A product for Hong Kong or Guangdong listeners should start with those three.

## Which one gives a brand its own Chinese voice?

[MiniMax's Speech-02 announcement](https://www.minimax.io/news/speech-02-series) describes voice cloning. [Fish Audio's cloning guide](https://docs.fish.audio/features/voice-cloning) asks for clips of at least 10 seconds each, and [its S2 page](https://fish.audio/s2/) names Chinese among the languages S2 Pro supports. [ElevenLabs](https://elevenlabs.io/docs/overview/capabilities/voices) describes cloned voices beside its Voice Library and voices designed from a text prompt.

Clone a voice only with the speaker's written consent, and read each vendor's terms on commercial use of a clone before you ship. Paxa Labs offers no voice cloning. A new voice comes from [voice design](https://paxalabs.com/voice-design), which builds one from a written description with no recording, and the designed voice reads Chinese, English, and Thai. [Voice design and voice cloning compared](https://paxalabs.com/guides/voice-design-vs-voice-cloning) sets the two side by side.

## What should you listen for in a Mandarin AI voice?

A language list says a vendor accepts Chinese text. How a voice reads your script is something to hear. Put these into one test script and run it through each candidate.

- Neutral-tone syllables. The second syllable of 妈妈, 东西 (a thing), and 明白 is light and short. A voice that gives it a full tone sounds stiff.
- Characters with two readings. 行 is háng in 银行 and xíng in 行走. 重 is zhòng in 重要 and chóng in 重新. 长 and 还 change the same way.
- Tone changes in running speech. 你好 is said with a rising first syllable, and 一 and 不 shift tone before certain syllables, as in 一个 and 不对.
- Numbers and dates. Listen to 2026年3月5日, an amount such as ¥1,280.50, a percentage, and a phone number read digit by digit.
- English inside the sentence. A line such as 请打开 Wi-Fi 设置 or a brand name written in Latin letters shows how the voice moves between the two languages.

## Jobs and vendors at a glance

| Job | Vendors whose documentation states the fit |
| --- | --- |
| Apps and websites in Mandarin | Amazon Polly, Azure AI Speech, Botnoi Voice, Google Cloud, MiniMax, Paxa TTS Flash v2 |
| Narration with captions | Amazon Polly, Azure AI Speech, Cartesia, Narakeet, WellSaid Labs, Paxa TTS Flash v2 |
| Voice agents and phone lines | Cartesia, ElevenLabs, Fish Audio, MiniMax, Murf, Paxa TTS Flash v2 |
| Chinese with another language in one voice | Azure AI Speech, ElevenLabs, Paxa TTS Flash v2 |
| Cantonese | MiniMax, Narakeet, WellSaid Labs |
| A cloned voice | ElevenLabs, Fish Audio, MiniMax |

## What should you try next?

- Hear 9 Paxa TTS Flash v2 voices read your own Chinese line with no account on [the free text to speech page](https://paxalabs.com/tts).
- Play the 2 Mandarin voices on [the Chinese voices page](https://paxalabs.com/chinese-voices), or all 50 on [the voices page](https://paxalabs.com/voices).
- Paste a full script into [the Speech tool](https://paxalabs.com/app/speech).
- Read about the product on [the text to speech page](https://paxalabs.com/text-to-speech).
- Compare published rates on [the text to speech API comparison](https://paxalabs.com/compare/text-to-speech-apis).
- See what the lab shipped in [Mandarin Chinese and six new voices](https://paxalabs.com/blog/mandarin-chinese-and-six-new-voices).
- Teaching Mandarin? [The Chinese learning page](https://paxalabs.com/chinese-learning) covers listening material.

## Common questions

### Which Chinese text to speech is best?

The one that fits your job. Find your job above, then test the two or three vendors it names on your own script, on the device your listeners use.

### Is Mandarin text to speech the same as Chinese text to speech?

Usually. Most vendors mean Mandarin when they write Chinese, with codes such as zh-CN or cmn-CN. Cantonese and Taiwanese Mandarin are listed separately where a vendor offers them. Paxa TTS Flash v2 speaks Mandarin from Simplified Chinese text.

### Can one AI voice speak Chinese, English, and Thai?

Yes. Every Paxa TTS Flash v2 voice reads all three, including a sentence that mixes them. A vendor whose voices each speak one language gives a brand a different voice per language.

### Does ElevenLabs support Mandarin?

Yes. [Its models page](https://elevenlabs.io/docs/overview/models) lists Mandarin Chinese among the 90+ languages of its newest speech models. [Our ElevenLabs comparison](https://paxalabs.com/compare/elevenlabs) sets the two services side by side.

### Is there a free Chinese text to speech?

[The free text to speech page](https://paxalabs.com/tts) reads a Chinese line aloud in 9 voices with no account. A new Paxa account also comes with free credits and no card, for personal use and evaluation.

### Where are the prices?

On [the text to speech API comparison](https://paxalabs.com/compare/text-to-speech-apis), which lists each vendor's published rate beside ours and is re-checked every month.
