Open source podcast generation pingtai is here, MoonCast, bilingual conversations are more natural!

MoonCast is an open-source, interactive voice synthesis model that generates a natural, Chinese-British, bilingual AI podcast by means of a few seconds of human voice samples to say goodbye to mechanical feelings; technological breakthrough I: a podcast using LLM to extract information to produce a summary and to create “human taste” with words such as fillers, response words and natural cartons; technological breakthrough II: an ultra-long audio generation of more than 10 minutes using a 2.5 billion-billion parameter model, large-scale training data and 40k context length, through three stages of training and short-scale self-regression audio reconstruction。