SOLIPSIUMAlpha
SOLIPSIUM Super resolution audio engine
Audio upscaler
Solipsium rebuilds the treble of speech saved at 16 kHz from the sound below 8 kHz, and labels what it adds as inferred. See where your file stops, then hear the first 20 seconds at 48 kHz.
Drop a 16 kHz recording here
Solipsium checks in your browser where its sound stops and what else is wrong, without uploading it. It can then repair the first 20 seconds for you to hear, with no account.
WAV, MP3, M4A, video and more, up to five minutes.
After signing up, drop the same file into the editor to repair all of it.
Then drop the same file into the editor. Scans and short repairs are free.
How it works
A file saved at 16 kHz holds sound up to 8 kHz, so speech in it sounds dull. The check in your browser names where your file's sound stops, and the audio spectrum check draws it as a picture.
When the sound stops between about 8.3 and 11 kHz and nothing else is found, AP-BWE's 16 kHz model rebuilds the treble. A steady noise floor is cleaned first with Full-band speech cleanup. The try's result then reads: “The recording seems to have been saved at 16 kHz, so nothing is left above 8 kHz. The treble is rebuilt from the sound below it, which is kept as it was. What is added is inferred.”
Solipsium's server decodes every recording at 48 kHz, and the repair comes back at that rate. The rebuild runs on the server's processor, so the free try does it too. The try plays Original and Repaired at the same level and offers its 20 seconds as a WAV download.
Sound that stops under about 5.4 kHz is a phone line, with a rebuild of its own on the muffled audio page. Sound that stops between 5.4 and 8.3 kHz is kept as it is.
Prism-SR is Solipsium's own model for muffled speech. Coming soon.
Questions
No. A 16 kHz file resampled to 48 kHz makes a bigger file that still stops at 8 kHz. The sample rate converter changes the rate in your browser when a program asks for a different one.
Drop it in the box above. The check names where the sound stops, and a file saved at 16 kHz stops at or a little above 8 kHz.
It is an estimate. The model predicts the sound above 8 kHz from the sound below it, and the result labels what it added as inferred. Everything below 8 kHz is your file's own sound.
The 16 kHz rebuild needs sound that stops between about 8.3 and 11 kHz, with nothing else found or only steady noise. When clipping, dropouts or hum are found as well, those are repaired and nothing is rebuilt. Sound that reaches past 11 kHz does not count as muffled. Sound that stops under 8.3 kHz is a phone line, with its own rebuild, or is kept as it is. The try lists its reasons beside the result.
Solipsium's free try is built for speech. For music, a free account's editor can run AudioSR on a passage of up to 60 seconds, and what it adds is inferred.
Not for the check, which runs in your browser. Pressing Repair the first 20 seconds sends those 20 seconds; when your browser could not open the file, or it runs over five minutes, the file itself goes to Solipsium's server, so the try is offered for it only up to 25 MB. Solipsium keeps only the first 20 seconds and deletes them within an hour. If you are signed in, a file you drop or choose is uploaded to your account and opens in the editor instead. The privacy notice has the rest.
Free audio tools
Pages for other narrow recordings, and tools that read where the sound stops.
Rebuild the treble of a phone-line recording, labelled inferred, and hear 20 seconds free.
Change a recording's sample rate in your browser, for a video editor or a phone system.
See a spectrogram of a file and where its sound stops, made in your browser.
See where a song's sound stops and whether it clips, in your browser.
Every free tool on one page.
A free account's editor runs the same Suggested repair over recordings up to 30 minutes long.
Free account, no card.