Ant group releases open-source speech models Ming-3.0 for text-to-speech, audio understanding, and speech recognition

On October 20, 2026, Ant Group announced the open-sourcing of its Ming-3.0 speech series, which includes models for text-to-speech (TTS), audio understanding and dialogue, and speech recognition. The series comprises five models, allowing developers to choose based on their specific tasks or combine different capabilities into their workflows. The Ming-3.0-tts model enables users to describe target voice characteristics through text or reference audio, with a unique director mode for specifying performance nuances, such as emotional shifts, emphasis, and pacing. The Ming-3.0-audio-flash and tiny models support advanced audio processing, including transcription, summarization, and sentiment analysis of long audio files, while preserving contextual elements like speaker identity and environmental cues. These models are already integrated into various Ant applications, including the Afu app, medical voice cloning, and security verification. Developers can access the models via Ling Studio, API services, and platforms like Hugging Face and ModelScope.