Most old recordings have one of three problems: steady hiss or hum, uneven loudness, or a format nothing modern reads. AI audio restoration fixes the first two and converts the third, on your own computer, with nothing uploaded.
The honest version of “restore with AI” is three tools: take out the steady noise, even out the volume, and save in a format your device reads. That set clears tape hiss, mains hum, a muffled phone call, and a voice recorded too quiet. It does not rebuild a waveform that was clipped, and it is tuned for speech, not music.
This guide uses Gleam Audio, which purplecandy makes. It runs a speech-trained model locally, works offline, and is free for personal use (up to 20 files a week). The controls below are from version 1.1.0.
What you need
- The recording in a common format: wav, mp3, m4a, flac, ogg, opus and more.
- Gleam Audio for macOS, Windows or Linux. It brings its own ffmpeg, so there is nothing else to install.
1. Install the app
Download the file for your computer from the Gleam Audio page.
- macOS: open the disk image and drag the app to Applications.
- Windows: run the setup file.
- Linux: one command installs it, no sudo needed:
curl -fsSL https://static.gleamaudio.com/gleam-audio/releases/latest/download/install.sh | bash2. Add the recording
Drop the files on the window, or click Open Files…. You can add a whole session at once; every tool except Cut and Split runs on the full list.
3. Run Remove noise
Pick Remove noise in the sidebar (it’s first). The Strength slider decides how much of the original room sound stays: lower it to keep more of the room.
At the bottom, choose where results go: next to the originals (the default) or a folder you pick. Then click Run. New files get the old name plus what was done, like interview-clean.wav.
4. Even out the loudness
A restored file can still be too quiet or uneven. Run Enhance on the cleaned files and pick the destination: music streaming, podcasts, audiobooks or broadcast. It sets the loudness each one expects and drops low rumble too. (What “loudness” means here: see LUFS.)
5. Convert for your device
Convert writes mp3, wav, m4a, flac, ogg or opus at the quality you choose. Pick the one your device expects: mp3 for old players, m4a for audiobook apps, wav or flac for an archive.
When it won’t help
- Music. Remove noise is tuned for speech: interviews, podcasts, lectures and calls. On music it can take out more than you want, so try a lower Strength or skip it.
- Clipping and distortion. Noise removal takes noise out; it can’t rebuild a waveform that was clipped while recording.
- Crackle and dropouts. Remove noise targets steady sounds: fans, hum, traffic, room tone. Click-and-crackle damage from a worn tape is not steady, so it stays.
Takeaways
- The whole task is three tools: Remove noise, Enhance, Convert.
- Keep Strength lower when anything in the file is not voice.
- Originals are never changed, so restoration is reversible.
- Everything runs offline; nothing is uploaded.