MiniMax launches Speech 2.8 with sound tags and voice cloning
MiniMax introduced Speech 2.8, a synthetic speech model featuring native sound tags, 10-second voice cloning, cleaner audio and cross-lingual updates.
คำตอบโดยย่อ
What did MiniMax announce with Speech 2.8?
MiniMax introduced Speech 2.8, a synthetic speech model with native sound tags for breaths and hesitations, voice cloning from a 10-second sample, and processing intended to reduce noise and distortion. The company also says it improved cross-lingual speech for Mandarin and Japanese and made the model available through its platform and Audio product.
ประเด็นสำคัญ
- MiniMax introduced Speech 2.8 on January 23, 2026.
- The model supports native sound tags for fillers, breaths, laughter and other vocal details.
- MiniMax says Speech 2.8 can clone a voice from a 10-second audio sample.
- The company says it reworked audio processing to reduce background noise and synthetic distortion.
- Cross-lingual improvements initially cover the Mandarin-Japanese language pair.
What MiniMax announced
MiniMax introduced Speech 2.8, a synthetic speech model that the company says adds native sound tags, higher-fidelity voice cloning, revised audio processing and improved cross-lingual performance. MiniMax announced that the model is live and directed users to its Open Platform documentation and MiniMax Audio product.
The company presented the release as an effort to make generated speech reproduce more of the pauses, breaths and hesitations found in human conversation. It also provided audio demonstrations for sound tags, voice cloning, noiseless output and cross-lingual speech.
Native sound tags
Speech 2.8 introduces native support for sound tags intended to reproduce colloquial and nonverbal elements. MiniMax says the model can handle fillers such as “um,” “uh” and “ah” while preserving rhythm, pitch and pauses.
The demonstration text also uses tags for a chuckle, breathing, throat clearing and laughter. According to MiniMax, these controls are designed to add hesitation, emphasis and other conversational details to generated speech.
Voice cloning from a short sample
MiniMax says it optimized Speech 2.8's feature-extraction process for voice cloning. The company says the model can use a 10-second recording to capture characteristics including vocal texture, breathiness and speaking pace.
The announcement includes an English demonstration comparing original audio with a cloned result. MiniMax describes the example in terms of conversational delivery, changing cadence and a warm mid-range vocal quality.
Audio processing changes
MiniMax says it re-engineered the model's processing engine to reduce background noise and synthetic distortion. The company describes the intended output as clear audio resembling a professional studio recording.
A separate demonstration accompanies this part of the announcement. Its sample narration contains quiet descriptive prose about a forest, wind and dawn, providing an example of the model's output without background noise.
Cross-lingual updates
Speech 2.8 also includes changes to cross-lingual generation. MiniMax says it addressed “accent bleed,” which it describes as unnatural tones and pronunciation changes that can occur when generating speech across languages.
The first stated language pairing is Mandarin and Japanese. MiniMax says the update is intended to make voices in those languages sound like native speakers. The company also says it plans to extend this work to additional languages, but it does not identify them or provide a release schedule.
Availability
MiniMax says Speech 2.8 is now live. The announcement provides access points through the MiniMax Open Platform's model documentation and the MiniMax Audio product.
Source: MiniMax's “MiniMax Speech 2.8: Breathing Life into AI Voice” announcement, published January 23, 2026.
คำถามที่พบบ่อย
- What is MiniMax Speech 2.8?
- MiniMax describes Speech 2.8 as its new synthetic speech model. It includes native sound tags, voice cloning, revised audio processing and cross-lingual improvements.
- How much audio does Speech 2.8 require for voice cloning?
- MiniMax says the model can clone a voice from a 10-second sample, capturing characteristics including texture, breathiness and speaking pace.
- Which languages received cross-lingual improvements?
- MiniMax says the initial cross-lingual update addresses its Mandarin-Japanese pair, with changes intended to reduce unnatural tones and pronunciation shifts.
- Where is Speech 2.8 available?
- MiniMax says Speech 2.8 is live and directs users to the MiniMax Open Platform and MiniMax Audio.