TMTPOST — Social platform Soul App’s artificial intelligence team open-sourced an end-to-end multi-speaker conversation transcription model on Wednesday. The model, named SoulX-Transcriber, was developed in collaboration with Northwestern Polytechnical University’s ASLP@NPU research group and Moonstep AI. It is specifically engineered to handle long-form conversational audio and complex multi-speaker social scenarios. By operating on an end-to-end architecture, the speech understanding system directly processes audio inputs to generate structured outputs. These final results automatically incorporate precise timestamps, distinct speaker labels, and fully transcribed text sequences.