Ant Ling open-sources five Ming-3.0 speech models across TTS, audio and ASR

Ant Ling announces it has open-sourced five Ming-3.0 speech models across three families: Ming-3.0-tts, a 0.5B model for voice design and paralinguistic cues, and a Tiny 7.9B A1.3B model for long-form speech; and Ming-3.0-asr, a Tiny 8.6B A2B model for dialects and context.

The announcement is on X and the report gives no further benchmark results.

A developer, Meng Shao, says he plans to run the 0.5B Ming-3.0-tts model locally because of its small size. He plans to build a book-reading assistant for his son and a German dubbing pipeline combining TTS, ASR and video understanding, with a working version due in a week and open-sourcing to follow. He has not yet released that project.