MiniMax Speech-2.6 on Novita AI: Next-Gen TTS Model for Voice Synthesis

Novita AI has expanded its speech generation suite with full support for the MiniMax Speech-2.6 series, featuring four advanced variants. This release delivers stronger multilingual expressiveness, more accurate voice replication, and broader coverage with 40 languages, making it ideal for real-time applications and long-form audio generation alike.

In this article, we’ll introduce what’s new in Minimax Speech-2.6, explain its features and key highlights, and show you how to get started with the API on Novita AI.

What is Minimax Speech-2.6?

MiniMax Speech 2.6 is the newest generation of speech technology, delivering sweeping enhancements such as ultra-low latency, improved format compatibility, and smoother, more lifelike voice output, making it ideal for powering natural and responsive Voice Agent experiences. The series includes four specialized variants, including MiniMax Speech-2.6-hd Text to Speech, MiniMax Speech-2.6-hd Async Long TTS, MiniMax Speech-2.6-turbo Text to Speech, and MiniMax Speech-2.6-turbo Async Long TTS, each designed to meet different application needs.

Minimax Speech-2.6: HD vs Turbo

Feature	Minimax Speech HD	Minimax Speech Turbo
Audio Quality	Ultra-realistic, studio-grade clarity	High-definition, but less expressive
Processing Speed	Higher latency, quality prioritized	Low latency, instant generation
Cost	Higher cost due to fidelity	Cheaper than HD
Emotions Support	Advanced emotion expression	Emotion support, slightly less nuanced
Best Use Cases	Audiobooks, media, narration	Chatbots, assistants, real-time apps
Parameter Controls	SSML, phoneme control, advanced options	Fast TTS, emotion, multilingual, API-friendly

Minimax Speech-2.6: Sync vs Async

Mode	Description	Best Use Cases
Synchronous	Converts text to speech instantly in real time	Live voice assistants, chatbots
Asynchronous	Processes text separately; results delivered later	Audiobooks, batch jobs, announcements

Minimax Speech 2.6: Key Highlights

1. Low Latency, High Responsiveness: Enabling Effortless Real-Time Interaction

The entire audio generation pipeline has been thoroughly reengineered to deliver an end-to-end latency below 250 milliseconds, reaching one of the highest standards in the industry. This breakthrough ensures that even in scenarios demanding instant feedback, such as real-time voice conversations or interactive assistants, audio generation remains smooth and uninterrupted. The result is a far more seamless and natural flow of communication, allowing every exchange to feel immediate and human-like.

2. Smarter Processing of Specialized Formats: Enabling Fluid, Accurate Information Delivery

Speech 2.6 introduces intelligent handling for a wide range of specialized text formats across multiple languages, including URLs, email addresses, phone numbers, dates, and currency expressions. The system can now interpret and vocalize these formats directly, without relying on external pre-processing steps or additional scripting. This makes it especially effective when paired with large language models or applications that manage dynamic, real-time data. By ensuring that every piece of information is read correctly and naturally from the outset, Speech 2.6 provides a more coherent, efficient, and human-sounding delivery of complex content.

3. Enhanced Naturalness: Delivering Authentic and Expressive Voices

Beyond its improvements in prosody and vocal tone, Speech 2.6 introduces the new Fluent LoRA technology, designed to achieve greater smoothness and realism in generated speech. Building on the high-fidelity voice cloning foundation of Speech 2.5, this version captures subtle features such as individual accents, rhythm, and speech habits with remarkable precision. Even when the source recordings include imperfect samples or non-native pronunciations, Fluent LoRA can faithfully reproduce the voice’s timbre while generating speech that is both fluent and expressive. This advancement allows Speech 2.6 to bring out the natural personality and clarity of every voice, making digital speech more engaging and emotionally resonant than ever before.

Minimax Speech 2.6: Applications

Model Variant	Type	Key Strengths	Ideal Applications
MiniMax Speech-2.6-HD Text-to-Speech	High-Definition Real-Time TTS	Studio-grade clarity, expressive tone control, accurate emotion rendering	Premium virtual assistants, audiobooks, podcasts, and digital avatars where naturalness and vocal richness matter
MiniMax Speech-2.6-HD Async Long TTS	High-Definition Asynchronous Long-Form TTS	Stable, high-quality generation for extended content, low distortion over long durations	E-learning narration, long-form storytelling, video voiceovers, automated news reading
MiniMax Speech-2.6-Turbo Text-to-Speech	Fast Real-Time TTS	Ultra-low latency, lightweight for rapid response	Interactive voice agents, live customer support bots, real-time communication tools
MiniMax Speech-2.6-Turbo Async Long TTS	Fast Asynchronous Long-Form TTS	Optimized for quick batch synthesis of longer texts	Mass content generation, large-scale dubbing, fast audiobook or media production pipelines

How to Use Minimax Speech-2.6 for Quick Voice Cloning on Novita AI?

Novita AI provides a REST API for voice cloning with Minimax Speech-2.6. MiniMax Speech-2.6 starts at $60 per 1M characters for the Turbo model and $100 per 1M characters for the HD model on Novita AI. You can get started in just a few simple steps using the API guide below.

​Step 1: Set Parameters

Header

Header	Type	Required	Meaning / Description
Content-Type	string	Yes	Specifies the media type of the request body. Use application/json.
Authorization	string	Yes	Bearer token for API authentication. Format: Bearer {API Key}. Example: Bearer sk-xxxxxx

Body

Parameter	Type	Meaning / Description
speed	number	Range: [0.5, 2], default is 1.0.
emotion	string	Controls the emotion of the synthesized speech. Currently supports 7 emotions: happy, sad, angry, fearful, disgusted, surprised, neutral.
text	string	Text (Sync: less than 10,000 characters / Async: less than 50,000 characters) to synthesize for preview. The result is returned as an audio URL.
model	string	Specifies the speech model for preview. Options: speech-2.6-hd, speech-2.6-turbo
voice id	string	Supports both system voices (ID) and cloned voices (ID). The available system voice IDs such as: Wise_Woman, Friendly_Person, Inspirational_girl, Deep_Voice_Man, Calm_Woman…

Step 2: Get API Key

Step 3: A Python Example

import requests

url = "https://api.novita.ai/v3/minimax-speech-2.6-hd"

payload = {
    "text": "<string>",
    "voice_setting": {
        "speed": 123,
        "vol": 123,
        "pitch": 123,
        "voice_id": "<string>",
        "emotion": "<string>",
        "text_normalization": True
    },
    "audio_setting": {
        "sample_rate": 123,
        "bitrate": 123,
        "format": "<string>",
        "channel": 123
    },
    "pronunciation_dict": { "tone": [{}] },
    "timbre_weights": [
        {
            "voice_id": "<string>",
            "weight": 123
        }
    ],
    "stream": True,
    "language_boost": "<string>",
    "output_format": "<string>",
    "voice_modify": {
        "pitch": 123,
        "intensity": 123,
        "timbre": 123,
        "sound_effects": "<string>"
    }
}
headers = {
    "Content-Type": "<content-type>",
    "Authorization": "<authorization>"
}

response = requests.post(url, json=payload, headers=headers)

print(response.json())

Frequently Asked Questions

Novita AI is the All-in-one cloud platform that empowers your AI ambitions. Integrated APIs, serverless, GPU Instance — the cost-effective tools you need. Eliminate infrastructure, start free, and make your AI vision a reality.