Audio Restoration
Audio restoration that treats the defect, not the recording
Euphona finds what is actually wrong with a recording — noise floor, clipping, clicks, too much room — and treats each defect with the tool that suits it.
Meters one minute per minute — a four-minute track costs four minutes of your allowance.
How it works
What it measures, what it changes
And what stays yours to decide. Every stage is a parameter you can read back and overrule.
- 01
It builds a defect map
The analysis pass marks where the recording is damaged and how: broadband noise and its floor, hum and its harmonics, clipped sample runs, clicks and dropouts, and how much of the room is in the signal. Each one is a different problem with a different fix. Treating them as one problem is why blanket noise reduction sounds the way it does.
- 02
Each defect gets its own treatment
Learned denoise for broadband noise, notch-and-track for hum, DSP interpolation with a learned reconstruction for clipping and clicks, learned dereverb for room. They're applied in the order that does the least damage: declip before denoise, because a clipped sample teaches a denoiser the wrong noise profile.
- 03
You set how far it goes
Every stage has a strength control, and the defaults are deliberately conservative. The failure mode of restoration isn't leaving a little noise in — it's a take scrubbed so hard the performance went with it. You can hear each stage on its own before committing to the chain.
The specifics
What you would ask before trusting it with a record
Formats, limits, and the places it is honestly hardest. Nothing here is a range we have not measured.
- Broadband noise
- Learned denoise working from the noise profile the analysis measured on your recording, not from a generic idea of what noise sounds like. At low strengths it keeps the floor's character instead of gating to silence between phrases.
- Clipping
- Clipped runs are detected by their flat-topped sample sequences and reconstructed by interpolation constrained to the surrounding waveform. Digital clipping reconstructs well; analogue overload is distortion that got recorded, and the report is honest about the difference.
- Clicks, pops and dropouts
- Impulse detection with local reconstruction, sized to the defect — a 2 ms dropout and a vinyl click aren't the same repair. Every repair is listed in the report with its timestamp so you can check it.
- Hum and buzz
- Fundamental and harmonic tracking at 50 Hz or 60 Hz, following drift rather than assuming a fixed frequency. A static 50 Hz notch misses hum that wanders, and it can take a bass note with it too.
- Room and reverb
- Learned dereverb reduces the recorded space rather than removing it, and that's usually the right call. A completely dry voice recorded in a real room sounds wrong in a way listeners notice but can't name.
Questions
What people actually ask
- Can you fix a clipped recording?
- Digital clipping, largely — the shape of the missing peak is recoverable from what surrounds it. What was clipped by an analogue stage before the converter is distortion that was recorded as part of the signal, and no restoration recovers what was never sampled.
- Will removing noise make my vocal sound underwater?
- That artefact comes from applying more reduction than the noise justifies, usually with one blanket filter. Working from a measured noise profile at a conservative default is the answer to it, and the strength control exists so you can hear where the line is on your own recording.
- Should I restore before or after mixing?
- Before, on the individual takes. Once a noisy track has been summed into a mix, the noise is correlated with everything else and the same repair does far more collateral damage.
- What about room sound I cannot get rid of?
- Dereverb reduces it and cannot delete it. The honest position is that a bad room is the one recording problem processing improves least, which is why it is worth the fifteen minutes of moving the microphone before the take.
Try it
Put your own track through it
The free plan runs the whole engine on your whole song and plays you the result, matched for loudness against the original. No card. You pay when you want the file — one track at a time, or a plan if you are making a lot of them.
Who it’s for