Solipsium
Solipsium provides a browser-based audio repair engine that automatically scans recordings, identifies issues like dropouts, noise, hum, and missing treble, and applies targeted fixes—including its Prism-PLC model for gap filling—while preserving the original file. Users can review each repair with timestamps, blend or undo changes, and export the cleaned audio. The service is currently free during its alpha phase, requiring only a web browser and no additional hardware.
- Artificial Intelligence
- Creator / Creative Economy
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
Damaged audio recordings—whether from phone calls, field recordings, or legacy media—suffer from dropouts, background noise, hum, clipping, and missing high frequencies. Existing repair tools are either hardware-dependent, require desktop software, or apply blanket processing without explaining what was changed or why, leaving users with opaque results and no way to verify the fixes.
Solution
Solipsium is a browser-based audio repair engine that scans a recording before making any changes, identifies each specific problem with timestamps, and applies targeted repairs selected by a benchmark-driven recommendation system. The engine includes Prism-PLC, a model that fills short dropouts by reshaping the fill band by band from surrounding sound, and a separate AudioSR step that rebuilds missing treble, with all inferred or estimated content clearly labeled. Every change is listed with its time, and the original recording is preserved as an alternative the user can keep, blend, or undo. The platform runs entirely in a web browser with no hardware requirements, and during its alpha phase all repairs are free.
Target Audience
Primary users are podcasters, content creators, journalists, and anyone working with recorded speech who needs to clean damaged audio without specialized hardware or desktop software, as well as professionals seeking transparent, verifiable repair results.
Features
- Automatic pre-repair scan that diagnoses each problem (dropouts, hum, hiss, missing treble, noise, echo, clipping) with timestamps before any processing
- Prism-PLC model for dropout filling that reshapes the fill band by band from surrounding sound, built for speech and tested on 80 recordings with 75 measured better and none worse
- Separate AudioSR treble rebuilding step that runs on the GPU and marks all reconstructed high frequencies as inferred
- Transparent repair workflow: every change is listed with its time, and estimated or inferred content is explicitly labeled
- Original recording preserved as an alternative that can be kept, blended, or undone
- Suggested repair mode that runs the plan that measured best on the detected damage type, or manual tool selection per passage
- Runs entirely in a web browser with no additional hardware or software installation