MiniMax Speech 2.8 Voice Model
MiniMax Speech 2.8Audio Model
AI BriefingMiniMax has released its next-generation voice model, MiniMax Speech 2.8. Using native voice tag technology, it simulates fillers such as "um" and "ah," as well as breathing pauses in human speech, significantly improving conversational naturalness. The model supports high-fidelity voice cloning from a 10-second sample, accurately reproducing vocal tone and speaking pace while removing background noise and digital artifacts to deliver clean, studio-quality audio. It also improves cross-language performance, starting with the Mandarin–Japanese pair to address accent bleed-through and achieve pronunciation closer to that of native speakers.
Source: MiniMax: Blog · minimax.io