Audio Upscaler

How to upscale audio

1. Upload a file. MP3, WAV, M4A or FLAC, up to 10 minutes.
2. Press Upscale the audio. The price is on the button before it runs: 1 credit per minute of audio. New accounts get 5 credits after signing in with Google.
3. Wait a few minutes. It usually takes 1 to 3 minutes, longer when it is busy.
4. Compare and download. You get one file, Restored, next to your original. Click the two reels to switch between them and hear the difference, then download the restored file as a 320 kbps MP3.

If a job fails, your credits are refunded.

What an audio upscaler does

Audio gets dull when the high frequencies are missing. This happens with old recordings, low-bitrate MP3s, files squeezed for streaming, voice memos made with a weak microphone and tracks that were cut off at a low frequency to save space. The sound is muffled, flat, with no air around a voice and no shimmer on cymbals.

An audio upscaler adds those highs back. The model here is AudioSR, which was trained to predict the high-frequency detail that a recording should have had, then writes it into the file. The result is a brighter, more open version of the same audio. It does not change the notes, the words or the timing. It fills in the top end.

When it helps

• A low-bitrate MP3 that sounds dull next to a lossless copy.
• An old digitized recording, such as a cassette, a home video soundtrack or a radio capture.
• A voice recording that lacks clarity, such as a lecture or interview where consonants feel soft.
• A sample or loop you pulled from a low-quality source and want to use in a project.
• Audio from a stem split that sounds a little muffled.

If your file has audible chirps, watery artifacts or a "swirly" sound from heavy MP3 compression, try Restore Compressed Audio first. It is built for that problem.

When it will not help

Be realistic about this. An upscaler is a repair for one thing, missing highs.

• Distortion and clipping stay. A crunchy recording will be a brighter crunchy recording.
• Noise and hiss stay, and added highs can make hiss more noticeable. Use the Vocal Noise Remover for that.
• Room echo stays. Use the Reverb Remover.
• A weak mix or a poor performance can't be fixed by adding highs.
• A phone recording won't turn into a studio recording. It can sound clearer and less muffled, but the mic and room are still in the sound.

The highs the model writes are a good estimate, not a recovery of what was originally there. That is why results vary: a dull but clean recording improves the most, and a damaged one improves the least.

Tips for a better result

• Start with your best copy. If you have the WAV or a higher-bitrate version, use it.
• Fix the other problems first, if you can. Remove noise before you upscale so the added highs don't bring it forward.
• Listen on good headphones or speakers. The change is in the top end, which small speakers can hide.
• Compare before you download. Switching between the original and the Restored reel is the quickest way to judge.
• Don't run the output through it again. A second pass doesn't add more real detail.
• Match the tool to the source. A dull voice memo and a dull 1970s tape transfer both suit this page. A track full of digital crackle or a badly clipped file does not, so spend your credits where the top end is the real problem.