MiniMax Speech 2.8 text-to-speech API
speech-2.8-hd for fidelity and speech-2.8-turbo for speed, billed per character, one line of code through the OpenAI-compatible /v1/audio/speech endpoint.
MiniMax Speech 2.8
Speech 2.8 is MiniMax's latest text-to-speech generation in two flavors: speech-2.8-hd for high-fidelity, expressive narration, dubbing and ads, and speech-2.8-turbo for low-latency, lower-cost real-time dialogue, notifications and bulk generation. Both cover Chinese, English, Japanese, Korean and more with a large catalog of built-in voices.
Hear it first
The same sentence rendered by both models, generated through this relay with no post-processing. Play them back to back and pick the one your project needs.
Voices used: English_expressive_narrator for the narration pair and English_CalmWoman for the presenter pair.
MiniMax TTS API pricing
| Model | Profile | Price |
|---|---|---|
| speech-2.8-hd | High fidelity | $1.00 per 10k characters ($100.00 per 1M) |
| speech-2.8-turbo | Fast | $0.60 per 10k characters ($60.00 per 1M) |
Billing is per input character, and Chinese or English characters count the same. A 2,000-character article costs about $0.06 with turbo and $0.10 with hd.
Quick start (OpenAI-compatible /v1/audio/speech)
curl
curl https://yiduochan.com/v1/audio/speech \
-H "Authorization: Bearer $YIDUOCHAN_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "speech-2.8-turbo",
"input": "Hello, welcome to MiniMax text to speech.",
"voice": "English_expressive_narrator",
"response_format": "mp3"
}' --output speech.mp3Python (OpenAI SDK)
from openai import OpenAI
client = OpenAI(base_url="https://yiduochan.com/v1", api_key="YIDUOCHAN_API_KEY")
with client.audio.speech.with_streaming_response.create(
model="speech-2.8-hd",
voice="English_expressive_narrator",
input="Hello, welcome to MiniMax text to speech.",
response_format="wav",
) as response:
response.stream_to_file("speech.wav")Parameters
| Parameter | Description |
|---|---|
model | speech-2.8-hd or speech-2.8-turbo |
input | Text to synthesize, billed per character |
voice | A MiniMax voice id such as English_expressive_narrator, male-qn-qingse or female-shaonv |
response_format | mp3, wav, flac or pcm, default mp3 |
speed | Speaking rate from 0.5 to 2.0 |
metadata | Pass-through for MiniMax-specific options such as voice_setting.emotion or audio_setting.sample_rate |
Typical uses
- Audiobooks, podcasts and video voice-overs generated in bulk.
- Real-time speech for support bots and voice assistants.
- Localization: one script rendered in several languages and voices.
FAQ
How is the MiniMax TTS API billed?
Per input character: speech-2.8-hd costs $1.00 per 10k characters and speech-2.8-turbo $0.60 per 10k characters. Failed requests are not charged.
speech-2.8-hd or speech-2.8-turbo?
Pick hd for audio quality and expressiveness in dubbing and audiobooks; pick turbo for low latency and low cost in real-time dialogue and bulk jobs.
Which audio formats are supported?
response_format accepts mp3, wav, flac or pcm; the endpoint returns raw audio bytes you can save or stream.
Can I call it with the OpenAI SDK?
Yes. The endpoint mirrors OpenAI's /v1/audio/speech; set base_url to https://yiduochan.com/v1 and use a MiniMax voice id for voice.
Is voice cloning available?
Yes. Voice cloning and voice design are exposed under /v1/minimax/ — upload a sample to /v1/minimax/files/upload and call /v1/minimax/voice_clone, or describe a voice in words at /v1/minimax/voice_design. Both are billed per created voice. The gateway issues the voice_id itself, so a voice stays private to the account that made it; a voice_id sent in the request is replaced. MiniMax's built-in voices remain available to everyone.