Last update: June 30, 2026
MiniMax Speech 2.8 is a MiniMax text-to-speech model family announced on January 23, 2026, for generating natural, expressive AI voice output. It is built for use cases such as narration, product voiceovers, voice interfaces, multilingual audio, and approved cloned-voice workflows, with two main API variants: speech-2.8-hd and speech-2.8-turbo.
Quick Facts Table
| Item | Details |
|---|---|
| Model family | MiniMax TTS / AI voice generation |
| Main variants | speech-2.8-hd and speech-2.8-turbo |
| Main purpose | Convert text into expressive synthetic speech |
| Best for | Narration, storytelling, localization, voice agents, learning apps, product demos, and audio prototyping |
| Key features | Natural speech generation, sound tags, interjection tags, voice cloning, emotion controls, multilingual support, streaming support |
| Access routes | MiniMax API, MiniMax audio products, and selected third-party model providers |
| Important caveat | Pricing, latency, limits, voice availability, and deployment behavior can vary by provider, account setup, and integration route |
What Is MiniMax Speech 2.8?
MiniMax Speech 2.8 is a text-to-speech and AI voice generation model family from MiniMax. It turns written text into spoken audio, with emphasis on natural cadence, expressive delivery, controllable voice settings, and more human-like non-verbal sounds.
The official MiniMax announcement describes Speech 2.8 around native sound tag support, high-fidelity cloning, and studio-grade clarity, positioning it as a speech model focused on vocal authenticity rather than basic robotic narration.
For developers, MiniMax Speech 2.8 is most important as an API-accessible TTS option. The MiniMax Text to Speech HTTP documentation lists speech-2.8-hd and speech-2.8-turbo as selectable speech synthesis model versions, alongside earlier MiniMax speech models.
The model family is useful when plain text-to-speech is not enough. Instead of only generating a flat spoken version of a script, it can support expressive narration, pauses, emotional tone, pronunciation guidance, and interjection tags such as breaths or laughter. These controls make it more suitable for storytelling, AI companions, guided learning, customer support voice, and content production.
MiniMax Speech 2.8 HD vs Turbo
| Comparison point | speech-2.8-hd | speech-2.8-turbo |
|---|---|---|
| Quality focus | Higher emphasis on fidelity, clarity, and polished narration | Balanced quality with stronger speed and iteration focus |
| Speed / latency orientation | Better fit when quality matters more than response time | Better fit when generation speed or throughput matters more |
| Production use | Suitable for polished voiceovers, premium narration, demos, and brand audio | Suitable for product workflows, voice interfaces, iteration, and higher-volume generation |
| Narration | Strong choice for long-form storytelling, educational scripts, and carefully produced videos | Good choice for scripts where speed, repeat generation, and flexibility are priorities |
| Real-time or high-volume use | Use when audio quality justifies possible latency tradeoffs | Use when responsiveness and scale are important |
| When to choose | Choose HD if narration fidelity, voice quality, and expressive detail matter more | Choose Turbo if speed, iteration, or high-volume generation matters more |
MiniMax’s model introduction page describes speech-2.8-hd as focused on ultra-realistic quality with sound tags, while speech-2.8-turbo is described around speed and natural flow. The same MiniMax overview lists both with support for 40 languages and 7 emotions.
Third-party model pages follow a similar split. Cloudflare describes MiniMax Speech 2.8 HD as focused on studio-grade audio generation with emotion control, multilingual support, and voice cloning, while its Turbo page describes natural expressive speech with voice cloning, emotion control, 40+ language support, and faster speeds.
Key Features of MiniMax Speech 2.8
Natural speech and prosody
The main value of MiniMax Speech 2.8 is not simply that it reads text aloud. Its goal is to produce speech that feels less mechanical by handling pacing, rhythm, pauses, tone, and non-verbal expressions. In practical terms, this matters when a script needs to sound conversational, educational, dramatic, calm, promotional, or character-driven.
For creators, better prosody can reduce the amount of manual editing required after generation. For developers, it can make voice agents or learning tools feel more usable because the generated voice can better match the emotional context of the interaction.
Sound and interjection tags
Sound tags and interjection tags let the script include vocal behaviors directly inside the text. MiniMax’s own example includes inline tags such as (chuckle), (breath), (clear-throat), and (laughs) to shape pacing and conversational realism.
The MiniMax T2A HTTP documentation states that interjection tags are supported only when using speech-2.8-hd or speech-2.8-turbo. Supported examples include (laughs), (chuckle), (coughs), (clear-throat), (groans), (breath), (pant), (inhale), (exhale), (gasps), (sniffs), (sighs), (snorts), (burps), (lip-smacking), (humming), (hissing), (emm), and (sneezes). Always verify the current supported tag list in the live MiniMax documentation before production use.
Voice cloning
MiniMax Speech 2.8 can be used in voice cloning workflows. MiniMax’s announcement states that Speech 2.8 can capture vocal texture, breathiness, and speaking pace from a 10-second sample.
For API workflows, MiniMax also provides a Voice Clone endpoint. The documentation describes it as an API for rapid voice cloning and notes that unused cloned voices are deleted after 7 days. Uploaded clone audio must meet requirements such as accepted formats of mp3, m4a, or wav, a duration from 10 seconds to 5 minutes, and a file size not exceeding 20 MB.
Emotion and voice controls
MiniMax Speech 2.8 can be controlled through voice settings such as voice_id, speed, vol, and pitch. The standard T2A HTTP example in the MiniMax documentation shows these fields inside voice_setting, along with audio settings such as sample rate, bitrate, format, and channel.
Emotion settings can also matter for production workflows. For example, a learning app may need a calm explainer voice, while a game character may need a more excited or surprised delivery. These controls should be treated as creative direction tools, not as a replacement for audio review.
Multilingual support
MiniMax’s model introduction page lists 40 languages supported for both speech-2.8-hd and speech-2.8-turbo. The T2A HTTP documentation also includes language_boost, which can enhance recognition for specified languages and dialects, with auto available when the language type is unknown.
This makes MiniMax Speech 2.8 relevant for localization, multilingual learning tools, international product demos, and voice content that needs to move between languages. For high-value production, review pronunciation and accent behavior before publishing.
API and streaming options
The MiniMax T2A HTTP endpoint is designed for synchronous text-to-audio generation over HTTP. The official API example uses a POST request to /v1/t2a_v2, with an authorization token, JSON content type, a model value, text, language boost, voice settings, and audio settings.
The T2A HTTP documentation states that the text field must be less than 10,000 characters and recommends streaming output for texts over 3,000 characters. For real-time playback workflows, MiniMax also provides a T2A WebSocket API. This is important for long-form narration, because long scripts may need chunking, streaming, or asynchronous processing depending on the integration route.
Output formats and subtitles
MiniMax’s T2A HTTP documentation lists supported audio formats as mp3, wav, and flac for non-streaming, with streaming limited to mp3. It also includes subtitle controls, including sentence-level timestamps, word-level timestamps, and word-level streaming timestamps when streaming is enabled. For longer jobs, MiniMax also documents an async long-form TTS workflow.
These options are useful for video production, learning apps, accessibility workflows, dubbing previews, and audio players that need synchronized text.
Sound Tags and Interjection Tags
Sound tags are inline script controls that tell the model to add non-verbal or semi-verbal vocal behavior. Instead of writing only words, you can include cues such as:
(laughs)(chuckle)(breath)(sighs)(clear-throat)
These tags matter because human speech is not only vocabulary. A pause before a key sentence, a small breath between ideas, or a laugh after a casual line can change how the listener understands the script.
Pause markers are different. A marker such as <#0.5#> controls silence duration. It does not request a laugh, breath, or vocal expression. The MiniMax documentation describes pause control through markers in the form <#x#>, where x is the pause duration in seconds, and gives a valid range from 0.01 to 99.99.
Best practices:
Use tags sparingly. Too many tags can make narration feel over-directed.
Match tags to context. A (laughs) tag may fit a casual story, but it may weaken a serious product explanation.
Preview important audio. Small changes in punctuation, tags, and pauses can alter the performance.
Avoid over-tagging professional narration. In many business, education, and documentary scripts, clean pacing is more effective than frequent interjections.
API Example
Below is a simplified cURL example using speech-2.8-hd. Replace the token placeholder with your API key and verify endpoint details, supported values, and account-specific settings in the MiniMax documentation or your chosen provider’s API reference before deployment.
curl --request POST \
--url https://api.minimax.io/v1/t2a_v2 \
--header 'Authorization: Bearer YOUR_MINIMAX_API_KEY' \
--header 'Content-Type: application/json' \
--data '{
"model": "speech-2.8-hd",
"text": "Welcome to the product walkthrough. (breath) In this guide, we will explain the workflow step by step.",
"stream": false,
"language_boost": "English",
"output_format": "hex",
"voice_setting": {
"voice_id": "English_expressive_narrator",
"speed": 1,
"vol": 1,
"pitch": 0
},
"audio_setting": {
"sample_rate": 32000,
"bitrate": 128000,
"format": "mp3",
"channel": 1
}
}'This example is intentionally simple. Production apps may also need error handling, retries, content validation, user consent tracking for cloned voices, storage handling for returned audio, subtitle options, and monitoring for latency or cost behavior.
Best Use Cases
YouTube narration: MiniMax Speech 2.8 can fit explainer videos, tutorial narration, list videos, documentary-style voiceovers, and channel localization. Use HD when voice quality is central to the brand. Use Turbo when you need faster drafts or frequent script changes.
Audiobooks and storytelling: The model’s expressive controls and interjection support can help with character-driven narration, short stories, children’s content, and serialized audio. Long-form work should be reviewed carefully for pacing, pronunciation, and consistency.
Product demos: Product teams can use TTS for onboarding videos, release explainers, demo walkthroughs, and internal training clips. A consistent synthetic voice can reduce production friction when scripts change often.
AI companions: Voice agents, character bots, and companion apps benefit from expressive delivery. Sound tags can help make short responses feel less flat, but should be used with restraint.
Learning apps: Language learning, guided lessons, pronunciation practice, and educational explainers can benefit from clear voices, multilingual support, and subtitle timing.
Customer support voice: MiniMax TTS can be used to prototype support flows, IVR messages, onboarding prompts, and self-service voice experiences. For customer-facing deployments, test clarity across devices and accents.
Localization and multilingual voice content: The model’s multilingual support makes it useful for adapting scripts into several languages. Native review remains important for pronunciation, cultural tone, and pacing.
Prototyping voice interfaces: Developers can use MiniMax Speech 2.8 to create voice prototypes before committing to a full voice stack, especially when testing UX, persona, and script structure.
Limitations and Things to Check
MiniMax Speech 2.8 should be evaluated like any production speech model: by the output quality, reliability, legal requirements, and integration behavior in your exact workflow.
Provider-specific limits can vary. MiniMax’s current rate limits page lists T2A for speech-2.8-turbo and speech-2.8-hd at 60 RPM, while third-party providers may wrap the model with different schemas, dashboards, rate limits, file handling, and billing rules.
MiniMax’s official pay-as-you-go page currently lists Text to Audio pricing at $60/M characters for speech-2.8-turbo and $100/M characters for speech-2.8-hd, plus separate charges for rapid voice cloning and voice design. Pricing can change by provider and account route, so verify the live MiniMax pay-as-you-go pricing before production planning.
Voice cloning requires consent. Do not clone a real person’s voice unless you have clear permission and rights for the intended use. Also check whether your platform requires disclosure, watermarking, or additional review.
Audio review is necessary before publishing. Sound tags, emotion settings, punctuation, and pronunciation controls can all change the final performance.
Latency can differ between HD and Turbo. HD is the better fit when quality has priority; Turbo is the better fit when speed and repeated generation matter more.
Long text handling needs planning. The MiniMax T2A HTTP documentation requires text to be under 10,000 characters and recommends streaming for text over 3,000 characters, so long scripts may need segmentation or streaming workflows.
Language and accent behavior should be checked with real scripts. Multilingual support does not remove the need for review by a fluent speaker.
Commercial rights and terms must be reviewed. Check MiniMax terms, provider terms, voice rights, content rights, and disclosure obligations before using generated or cloned voices commercially.
MiniMax Speech 2.8 vs Earlier MiniMax Speech Models
MiniMax Speech 2.8 is best understood as a newer Speech 2.x generation focused on more expressive control, with sound and interjection tags as a major practical difference. The API documentation lists earlier options such as speech-2.6-hd, speech-2.6-turbo, speech-02-hd, speech-02-turbo, speech-01-hd, and speech-01-turbo, but interjection tags are documented as supported only with speech-2.8-hd and speech-2.8-turbo.
That does not mean earlier MiniMax speech models are poor choices. Some products may already rely on them for stability, cost structure, voice compatibility, or legacy integrations. The stronger reason to choose MiniMax Speech 2.8 is when your workflow needs more expressive vocal control, sound tags, or a clearer split between HD-style quality and Turbo-style speed.
For teams with existing MiniMax TTS integrations, the decision should be practical: compare generated samples on the same scripts, the same voice IDs, the same audio formats, and the same target languages. The right model is the one that fits the production requirement, not only the one with the higher version number.
Who Should Use MiniMax Speech 2.8?
Use speech-2.8-hd if quality and narration fidelity matter more. This is the better default for polished YouTube voiceovers, brand explainers, audiobooks, storytelling, premium product demos, and audio where listeners will notice subtle artifacts.
Use speech-2.8-turbo if speed, iteration, or high-volume generation matters more. It is better suited for drafts, prototypes, dynamic product experiences, voice agents, and workflows where response time affects user experience.
Choose another route if you need a specific provider integration, a broader voice library, a different licensing structure, a particular compliance setup, or a model already embedded in your product stack. The best implementation depends on your latency needs, budget model, legal requirements, output format, and review workflow.
FAQ
What is MiniMax Speech 2.8?
MiniMax Speech 2.8 is a MiniMax text-to-speech model family for generating natural and expressive AI voice audio from written text. Its main API variants are speech-2.8-hd and speech-2.8-turbo.
What is the difference between MiniMax Speech 2.8 HD and Turbo?
speech-2.8-hd is better suited for high-fidelity narration and polished audio. speech-2.8-turbo is better suited for faster generation, iteration, and higher-volume workflows.
Does MiniMax Speech 2.8 support voice cloning?
Yes. MiniMax provides voice cloning workflows, and Speech 2.8 can be used with cloned voice outputs. Voice cloning should only be used with proper consent and rights.
What are sound tags in MiniMax Speech 2.8?
Sound tags are inline text cues that add vocal expressions or non-verbal sounds, such as (laughs), (chuckle), (breath), (sighs), or (clear-throat).
Can I use MiniMax Speech 2.8 for YouTube voiceovers?
Yes. It can be used for YouTube narration, explainers, tutorials, and localized video content. Review the generated audio before publishing and confirm commercial usage rights for your access route.
Does MiniMax Speech 2.8 support multiple languages?
Yes. MiniMax’s model overview lists 40 supported languages for both speech-2.8-hd and speech-2.8-turbo.
How do I access MiniMax Speech 2.8 through an API?
You can use the MiniMax Text to Speech HTTP API by sending a POST request with a model value such as speech-2.8-hd or speech-2.8-turbo, text, voice settings, and audio settings.
Is MiniMax Speech 2.8 suitable for real-time applications?
It can be suitable for interactive workflows, especially when using the Turbo variant or streaming where supported. Test latency with your exact script length, provider, region, and output settings.
What should I check before using cloned voices commercially?
Check consent, identity rights, commercial license terms, disclosure rules, provider policies, storage rules, and whether the cloned voice must be used or retained under specific account conditions.
Conclusion
MiniMax Speech 2.8 is a strong fit when you need expressive AI voice generation with practical controls for narration, sound tags, voice cloning, multilingual content, and API-based production. Choose speech-2.8-hd when audio fidelity is the priority, choose speech-2.8-turbo when speed and iteration matter more, and always validate the final output, usage rights, and provider-specific limits before publishing or deploying.
More on speech: see the MiniMax Audio hub for voices, voice cloning, text-to-speech pricing and which speech model to choose.