- The five-model family covers three speech tasks that developers can combine in a workflow.
- The audio-flash model card lists MIT licensing and Chinese and English as its primary languages.
Five models across three tasks
Ant’s Ling team has announced the open sourcing of its Ming-3.0 speech family, covering speech synthesis, audio understanding and dialogue, and speech recognition. Phoenix Technology reported the announcement on October 20, describing five models that developers can use separately or combine in a workflow.
The report says the text-to-speech series supports voice design from written descriptions or reference audio. A director mode lets creators specify emotion, emphasis and pacing. It also says the audio-flash and audio-tiny models can transcribe, summarize and analyze long recordings.
Model card and business uses
The inclusionAI model card describes Ming-3.0-audio-flash as a native audio-language mixture-of-experts model with 126 billion total parameters and 6.5 billion activated per inference step. It lists Chinese and English as primary languages and an MIT license.
Phoenix reports that Ant already uses the speech models in services including its Afu app and Alipay search. These uses are reported company deployments; the sources do not establish independent performance comparisons or the original upload date of every model.
Sources & context
Go to the original material. Company claims remain attributed to their sources.
01
