UniverSR Online:
Powerful GPU-powered Audio Restoration

UniverSR is a heavier restoration model designed for difficult degraded audio, including sound effects, music, voice, and narrowband material. AudioSR is lighter and still useful for classic high-frequency regeneration pipelines.

Use AudioSR Online for faster high-frequency regeneration, and UniverSR Online when the source is harder and the extra compute is worth it.

UniverSR is useful for degraded, compressed, narrowband, or old-sounding audio. It can help rebuild high-frequency detail and reduce harsh high-frequency artifacts such as hiss, brittle cymbals, sibilance, or missing brilliance.

It works on more than music: foley, retro game sounds, sound effects, voice, archival recordings, and other degraded audio can benefit.

However, it is not the best tool for every problem. Bass stems and low-end muddiness often need Neural Remix, EQ, or stem-level editing instead of high-frequency restoration.

Start from UniverSR Online, or compare restoration models in the restoration docs.

No, it does not just add noise. Unlike enhancers that layer white noise, the generative model reconstructs the clean signal that should be there by separating useful harmonic and transient structure from compression artifacts and other degradations.

For more examples, open Audio Restoration.

UniverSR is one of the most compute-heavy restoration models in Neural Analog. It uses more GPU time than standard upscalers, so longer UniverSR jobs are reserved for Pro users and above.

Use UniverSR Online for difficult music, voice, archival, and sound-effect restoration jobs, and see the UniverSR model card for technical details.

Processing time depends on the operation, selected model, input length, and queue load. Short jobs often finish in about a minute, while long tracks, batch jobs, and heavier models such as SAM Audio Large or UniverSR can take several minutes or more.

You can leave the page and come back later. Finished imports, restorations, masters, and stems stay in Studio, where you can reopen the project and download the result.

Yes, restoration can work on old codecs and low bit-depth audio after the file is decoded into an audio signal. Results depend on how much usable information remains in the source.

Try a short sample with Audio Restoration. For retro game and sound-effect material, see the UniverSR sound effects guide.

Yes. Support depends on the model. UniverSR, AudioSR, and FlashSR upscale low resolution audio to 48 kHz (super resolution). The 'Music Upscaler' restoration algorithm keeps 48 kHz sources at 48 kHz, and restores 44.1 kHz-or-lower sources at 44.1 kHz.

For the model-level details, read the UniverSR sample-rate guide or compare models in the restoration docs.

Yes, you can run restoration more than once. You can change the selected Source to run another pass on already restored audio, and batch imports support an 'Iterative restoration' toggle that selects the latest restored version of each file.

Do this in Studio by selecting the restored version as the source before running another restoration. Stop when the result improves less than the artifacts it introduces.