AudioShake Launches The Refinery for AI Audio Training Data

AudioShake launches The Refinery, a system that converts existing recordings into structured data for training audio AI models. It separates simultaneous speakers into individual tracks and isolates dialogue, music and background sounds without requiring original stems or session files.

The system preserves interruptions, overlapping speech, accents and background conditions from the source recording. It also assigns quality and confidence scores, allowing customers to sort audio into usable, fixable and unusable groups. The Refinery does not generate or reconstruct speech.

AudioShake says private versions of The Refinery have processed hundreds of millions of minutes of audio for major AI labs over the past year. Customers can process recordings through an API or deploy the system on premises so sensitive audio remains within their own environment. Applications include speech recognition, speaker identification, speech synthesis and conversational AI.