IT之家·· 2026-10-20PicksAI score68
Ant Ling opens source Ming-3.0 speech series models, covering speech synthesis, audio understanding and dialogue, and speech recognition
AI briefing
Ant Ling has open-sourced the Ming-3.0 speech series, including three series—speech synthesis, audio understanding and dialogue, and speech recognition—with five models in total, which developers can select by task or combine into workflows. Ming-3.0-tts's Voice Design supports describing the target timbre in text, and can also set the timbre with reference audio; director mode can specify how lines are performed, emotional changes, stressed words, and pacing.
Why it matters
Ming-3.0 splits speech synthesis, audio understanding, and recognition into five models, among which director mode can describe the performance style of lines in text.
Open original ithome.com