Voice Presets
Provides 17 or more built-in voice options spanning different genders, ages, and speaking styles, selectable via a dropdown input.
MiniMax
Minimax Speech 2.8 HD
MiniMax Speech 2.8 HD is a high-definition text-to-speech model developed by MiniMax, built on an autoregressive Transformer architecture with a Flow-VAE decoder. Instead of using traditional mel-spectrogram vocoders, it models speech in a learned latent space, which produces audio with natural cadence, proper intonation, and emotional depth. The model accepts up to 50,000 tokens of input text and was trained through January 2026. The model offers 17 or more expressive voice presets spanning different genders, ages, and speaking styles, along with support for natural interjections such as laughs, sighs, and gasps embedded directly in text. Users can control emotion, speed, volume, pitch, sample rate, bitrate, channel configuration, and output format. These features make it well suited for audiobook production, video voiceovers, podcast creation, e-learning narration, accessibility applications, and game development.
January 2026
50K context
N/A output
Voice Presets
Emotion Control
Natural Interjections
Audio Format Control
Speech Rate & Pitch
Custom Pronunciation
Model Overview
High-signal model metadata in a structured two-column overview table.
Provider
The entity that provides this model.
MiniMax
Input Context Window
The number of tokens supported by the input context window.
50K
tokens
Maximum Output Tokens
The number of tokens that can be generated by the model in a single request.
N/A
tokens
Open Source
Whether the model's code is available for public use.
No
Release Date
When the model was first released.
January 2026
Knowledge Cut-off Date
When the model's knowledge was last updated.
January 2026
API Providers
The providers that offer this model. This is not an exhaustive list.
MiniMax
Modalities
Types of data this model can process.
Text
Video
Audio
Code
What is Minimax Speech 2.8 HD
A fuller summary of positioning, capabilities, and source-specific details for Minimax Speech 2.8 HD.
MiniMax Speech 2.8 HD is a high-definition text-to-speech model developed by MiniMax, built on an autoregressive Transformer architecture with a Flow-VAE decoder. Instead of using traditional mel-spectrogram vocoders, it models speech in a learned latent space, which produces audio with natural cadence, proper intonation, and emotional depth. The model accepts up to 50,000 tokens of input text and was trained through January 2026.
The model offers 17 or more expressive voice presets spanning different genders, ages, and speaking styles, along with support for natural interjections such as laughs, sighs, and gasps embedded directly in text. Users can control emotion, speed, volume, pitch, sample rate, bitrate, channel configuration, and output format. These features make it well suited for audiobook production, video voiceovers, podcast creation, e-learning narration, accessibility applications, and game development.
Capabilities
What Minimax Speech 2.8 HD supports
Emotion Control
Allows setting the emotional tone of synthesized speech — such as happy or calm — to match the intended content context.
Natural Interjections
Supports embedding over 20 human sounds like (laughs), (sighs), and (gasps) directly in input text for lifelike delivery.
Audio Format Control
Exposes configurable parameters for sample rate, bitrate, channel configuration, and output format through dedicated select inputs.
Speech Rate & Pitch
Accepts numeric inputs to adjust playback speed, volume level, and pitch independently for fine-grained audio tuning.
Custom Pronunciation
Supports a custom pronunciation dictionary to handle brand names, acronyms, and specialized terminology with precise phonetic control.
Large Text Input
Accepts up to 50,000 tokens of input text in a single request, enabling long-form content like full audiobook chapters.
Pricing for Minimax Speech 2.8 HD
Primary API pricing shown in the same “quick compare” spirit as the reference page.
Input tokens
N/A
Per million tokens
Output tokens
N/A
Per million tokens
API Access & Providers
Places where this model is available, based on the synced detail-page metadata.
MiniMax
Configuration & Parameters
The configurable options currently documented for this model.
Voice
SelectVoice preset to use for speech synthesis.
Speed
NumberSpeech speed multiplier.
Volume
NumberVolume level.
Pitch
NumberPitch adjustment.
Emotion
SelectEmotional tone of the speech delivery.
Sample Rate
SelectAudio sample rate in Hz.
Bitrate
SelectAudio bitrate in bits per second.
Channel
SelectAudio channel configuration.
Format
SelectOutput audio format.
Language Boost
SelectBoost recognition for a specific language.
English Normalization
Toggle GroupImproves number-reading performance in English text (dates, currencies, etc.).
Supported Request Parameters
Parameters currently listed by OpenRouter or the local catalog for this model.
Resources & Documentation
Official model cards, release notes, docs, and other references synced from the source page.
Community discussion
What people think about Minimax Speech 2.8 HD
Minimax Speech 2.8 HD discussions are most active in r/HiggsfieldAI. The strongest match in this snapshot has 1 upvotes and 0 comments.
The Ultimate AI Text-to-Speech, Voice Swap, and Video Translation Tool: Higgsfield Audio
Imagine this: you have spent hours crafting the perfect AI visual. The lighting is right, the motion is smooth, the style is exactly how you envisioned it. Everything seems almost too perfect - except the audio sounds off.
Our latest, breakthrough update brings Higgsfield Audio that turns Higgsfield AI into a full-cycle AI content production platform. With three powerful new functions, Voiceover, Change Voice, and Translate, you no longer need to leave the platform to give your content a voice.
🔊 **Create your custom voice:** upload an audio file or record on the spot.
🔊 **Voiceover:** text-to-speech that supports 70+ languages with 21 voice presets as well as your own custom voices. Powered by Eleven v3 and MiniMax Speech 2.8 HD.
🔊 **Change voice:** reshape how any video sounds. 21 voice presets + your custom voices.
🔊 **Translate:** localise your videos into 10 supported world languages:
🇺🇸 English
🇨🇳 Chinese (Mandarin)
🇫🇷 French
🇮🇳 Hindi
🇮🇹 Italian
🇯🇵 Japanese
🇰🇷 Korean
🇧🇷 Portuguese
🇷🇺 Russian
🇹🇷 Turkish
With more coming soon! Any guesses which ones are joining the list? Drop a comment 😗
Check Higgsfield Audio in action 👉 [https://higgsfield.ai/cinema-studio](https://higgsfield.ai/cinema-studio)
FAQ
Common questions about Minimax Speech 2.8 HD
What is the maximum input length for MiniMax Speech 2.8 HD?
The model supports a context window of 50,000 tokens, which allows for long-form content such as full chapters or extended scripts in a single request.
What audio output formats and quality settings are available?
Users can configure sample rate, bitrate, channel (mono or stereo), and output format through dedicated select inputs, giving full control over the final audio file.
Can I control how the voice sounds beyond just selecting a preset?
Yes. In addition to choosing from 17 or more voice presets, you can adjust speed, volume, pitch, and emotional tone, and embed natural interjections like (laughs) or (sighs) directly in the input text.
What is the training data cutoff for this model?
The model's training date is listed as January 2026.
What types of applications is MiniMax Speech 2.8 HD best suited for?
The model is designed for use cases that require high-fidelity, human-sounding audio, including audiobook production, video voiceovers, podcast creation, e-learning narration, accessibility tools, and game development.
More models from MiniMax
Continue browsing adjacent models from the same provider.