Ant Group open-sources Ming-3.0-tts speech models with voice design and fine control

Ant Group's inclusionAI team has open-sourced Ming-3.0-tts under the MIT license, with a 0.5B dense model and a 7.9B-total, 1.3B-active MoE tiny model.

The models offer voice cloning, voice design from text, over 100 Chinese and five English built-in voices, and fine control over emotion, speed, pitch and volume.

The README lists weak spots, including lower pitch-control accuracy and a weaker Korean error rate for the tiny model.