MiniMax Speech 2.8 Launches with Native Sound Tags and 10-Second Cloning
MiniMax has launched Speech 2.8, an AI voice model that introduces native sound tag support, high-fidelity voice cloning from a 10-second sample, and studio-grade clarity. The model is now available on the MiniMax Open Platform. According to the company, the update aims to make synthetic speech more human by modelling colloquial fillers like 'um' and 'uh', preserving natural rhythm and pauses. With just a short sample, the model captures a speaker's unique texture, breathiness, and speaking pace. This release also improves cross-lingual performance, starting with Mandarin-Japanese pairs, to reduce accent bleed. These features reportedly close the gap between AI and human voice. Since this is an official announcement, independent benchmarks are not yet available.
MiniMax Speech 2.8 Launches with Native Sound Tags and 10-Second Cloning
MiniMax released Speech 2.8, adding native sound tag support, high-fidelity cloning, and studio-grade clarity. The model is now live on the MiniMax Open Platform.
Key takeaway
MiniMax Speech 2.8 adds native sound tags and 10-second voice cloning, offering more natural and human-like AI speech with improved cross-lingual fidelity across languages.
What happened
According to MiniMax's official announcement, the company has released Speech 2.8, adding native sound tag support, high-fidelity voice cloning, and studio-grade clarity to its speech model. The release is now live on the MiniMax Open Platform.
The model can clone a voice from a 10-second sample, capturing the speaker's unique texture, breathiness, and speaking pace. It also models colloquial fillers like 'um' and 'uh' to preserve natural rhythm, and improves cross-lingual performance starting with the Mandarin-Japanese pair to reduce accent bleed.
Evidence
MiniMax released Speech 2.8 with native sound tag support, high-fidelity cloning, and studio-grade clarity.
Minimax · attributed
MiniMax released Speech 2.8, adding native sound tag support, high-fidelity cloning, and studio-grade clarity.
The model is now live on the MiniMax Open Platform.
Minimax · attributed
The model is now live on the MiniMax Open Platform.
With a 10-second sample, Speech 2.8 precisely captures unique texture, breathiness, and speaking pace.
Minimax · attributed
With just a 10-second sample, Speech 2.8 precisely captures your unique texture, breathiness, and even your specific speaking pace.
Cross-lingual performance is improved to eliminate accent bleed, starting with the Mandarin-Japanese pair.
Minimax · attributed
We're breaking down language barriers by eliminating the 'accent bleed' that often occurs in AI speech.
Why it matters
The ability to clone a voice from a 10-second sample and model natural conversational cues could lower barriers for creating realistic synthetic voices, potentially accelerating adoption in media and customer service, while also raising concerns about voice authenticity and misuse.
Limits and uncertainties
The announcement is from MiniMax alone and lacks independent verification or comparative benchmarks.
The scope of cross-lingual improvements is currently limited to Mandarin-Japanese pairs.
Practical implications
Developers can start integrating Speech 2.8 via the MiniMax Open Platform, leveraging the new sound tags and quick cloning for more natural voice interactions.
Builders should evaluate the model's performance and consider potential misuse of voice cloning.
What to watch
Look for third-party evaluations or benchmarks of Speech 2.8, and announcements of additional supported language pairs.
Monitor the platform's documentation for API specifics.
