AudioShake launches The Refinery to turn overlapping conversations into AI training data

AudioShake has launched The Refinery, a system that splits recordings into separate speaker tracks while separating dialogue from music and background sound. It works from existing recordings without regenerating speech and scores outputs for quality and confidence to help sort large audio archives.

• The Refinery splits conversations into individual speaker tracks while preserving overlapping speech
• It works from existing recordings without original sessions or separately captured stems
• On LibriCSS it reports 9.17% WER versus 37.75% for the tested MERL TF-Locoformer checkpoint
• Outputs are scored for quality and confidence to sort usable, fixable and unsuitable material