MiniMax Speech 2.6
Speech 2.6 focuses on responsive voice interaction and reading structured text aloud. Its launch introduced Fluent LoRA to improve fluency while retaining a cloned voice's timbre.
Voice identity and fluency are separate targets
MiniMax describes Fluent LoRA as a way to reduce disfluency in cloned speech without losing the source voice's timbre. The feature is intended to work even when the reference recording contains an accent or uneven delivery.MiniMax Speech 2.6 announcement
Check names and numbers with a human listener
Practical guidance. Use a sample script with actual date formats, amounts and names from the application. Correct text normalization matters more than an attractive demo voice when the listener must act on the information.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | speech-2.6-hdspeech-2.6-turbo |
| Variants | HD and Turbo remain listed in the current speech catalogue.MiniMax current model specifications |
| Text handling | The release describes direct handling of URLs, email addresses, telephone numbers, dates and money amounts.MiniMax Speech 2.6 announcement |
| Emotion controls | The API documents fluent and whisper options for Speech 2.6 HD and Turbo.MiniMax synchronous speech API |
Related models
- MiniMax Speech 2.8
MiniMax · MiniMax Speech
A speech-generation model with native breath and laughter tags, 40 languages and HD or Turbo serving options.
Launch date API · Provider product - MiniMax Speech 02
MiniMax · MiniMax Speech
A hosted speech model with 24 languages, seven emotions and HD or Turbo serving options.
Launch date API