MiniMax Audio

New

MiniMax Audio is MiniMax's browser studio and API for its Speech models, offering text to speech in 40 languages with sound tags, voice cloning from 10-second samples, voice design from text and long-form async synthesis.

minimax.io

MiniMax Audio

Explore features, practical uses and pricing below.

MiniMax Audio is the browser studio for MiniMax's speech and music models. Developers reach the same speech models through the MiniMax Open Platform API. MiniMax also makes the M-series language models and H3 video models. This listing covers the speech side: text to speech, voice cloning and voice design.

What the Speech 2.8 models add

MiniMax released Speech 2.8 in January 2026 in two versions, speech-2.8-hd and speech-2.8-turbo, the second tuned for speed and cost. The launch post says the main change is native sound tags: the model performs fillers and non-verbal sounds written into the script, such as (laughs), (breath) or (chuckle), to get closer to conversational pacing. The model list says the 2.8 and 2.6 models support 40 languages and seven emotions, while the older Speech-02 models cover 24 languages. An emotion parameter accepts values such as happy, sad, calm and surprised, though by default the model chooses an emotion from the text.

Cloning, designing and managing voices

Cloning starts from an uploaded recording. The voice clone guide asks for an MP3, M4A or WAV file between 10 seconds and 5 minutes long and up to 20 MB, plus an optional example clip under 8 seconds to improve similarity. You give the clone your own voice ID and then use it in any synthesis call. The API reference notes that a cloned voice not used within 7 days is deleted, so clone close to when you plan to record.

Voice Design creates a new voice from a written description instead of a recording. It suits characters or brand voices where no real person should be copied. System voices, cloned voices and designed voices can be listed or deleted through the voice management API.

Long-form and real-time synthesis

The synchronous endpoints take up to 10,000 characters per request over HTTP or WebSocket. A bidirectional WebSocket accepts text at any granularity which suits piping a language model's streamed output into speech. For audiobooks and long articles, the async long TTS guide allows up to 100,000 characters of direct text or per file, and up to 1,000,000 characters per task when you upload a zip. Each Chinese character counts as two. The API also accepts a pronunciation dictionary for tricky names and terms.

Narrating a course in two languages

An e-learning producer records two minutes of the lead instructor reading sample material, with the instructor's written consent, and clones the voice. She writes each lesson in English with a few (breath) tags at section breaks, adds the course's acronyms to the pronunciation dictionary, and sends the full script as an async task. She reuses the same voice ID for the Spanish version and has a native speaker review it, because accent carry-over between languages varies. MiniMax says its cross-lingual accent work started with the Mandarin-Japanese pair.

Who MiniMax Audio fits

MiniMax Audio suits developers who want expressive multilingual TTS and cloning priced per character, and creators producing narration, audiobooks or character dialogue in any of the 40 supported languages. Teams that mainly need a voiceover editor with collaboration and video tools may prefer Murf or ElevenLabs. The voice clone guide doesn't describe a consent-verification step, so commercial users should read MiniMax's terms on voice rights and keep their own consent records. Similar products are in our voice generators category.

How MiniMax charges for speech

The API pricing page lists speech-2.8-hd at $100 per million characters and speech-2.8-turbo at $60 per million, for both sync and async synthesis. Rapid voice cloning costs $1.50 per voice and Voice Design $3 per voice, each charged only when the voice is first used for synthesis. Audio Packages (Starter, Standard, Pro, Scale and Business) bundle monthly audio points, voice slots and rate limits as an alternative to pay-as-you-go. Pricing for the browser studio isn't shown on its public page, so check it inside the app.

Things people ask about MiniMax Audio

Why did my cloned voice disappear?

The API deletes cloned voices that aren't used for synthesis within 7 days.

Which model should I start with?

Turbo is the cheaper, faster option for drafts; HD costs more per character. Only the 2.8 models support the interjection tags.

Similar tools

Fish Audio generates expressive speech from text, clones voices from short clips and transcribes audio, with a 2M+ community voice library, inline emotion cues and a separate pay-as-you-go API for developers.
Dub recorded audio and video, refine translated dialogue and voices, or build localized speech workflows with CAMB.AI Studio and APIs.
AI Magicx combines model chat, image and video generation, voice and music tools, with creative exports and a separate app-building workflow.