According to Chinese media reports, the desktop version of ByteDance's generative AI tool 'Doubao' has recently launched an 'AI Podcast' feature. Users can upload PDF files or webpage links to generate a podcast program in the form of a two-person conversation.
The report indicates that this feature is based on a voice podcast model independently developed by the Doubao large model team. It achieves fast end-to-end conversion through a streaming architecture, with voice effects that closely resemble real human conversation in terms of tone, pauses, and fillers. However, the feature is currently only available to Doubao desktop users, while the Doubao app is undergoing limited testing and is expected to be fully available soon.