Ming-3.0 ships five open-weight audio models with open SOTA voice cloning
realmrfakename · x · 2026-10-19
ModelScope released Ming-3.0, a family of five open-weight audio models covering controllable speech, long-recording analysis, and context-aware transcription.
- Ming-3.0-tts-0.5B hits open-weight SOTA on Artificial Analysis Controlled Voice with a 1069 Elo rating using eight English reference voices; supports voice design from descriptions or references, multilingual synthesis, and controllable delivery.
- Ming-3.0-audio-flash transcribes, summarizes and answers questions grounded in long recordings, scoring 99.40 on BigBenchAudio S2T in the authors' comparison.
- Ming-3.0-asr-tiny accepts hotwords and domain terminology for business-context transcription.
Related event: Ant's inclusionAI Open-Sources Five Ming-3.0 Speech Models(6 posts)→