AUD · Audio tools

Audio Recorder

Ready
0:00
M
-60.0

Nothing is captured until you press record, and the level meter runs first so you can set the input before you do.

Shareable link

Settings are written to the URL as you change them. Nothing differs from the defaults yet.

Record from a microphone and get a file, with the encoder running while you talk rather than after you stop.

Set the level first

The meter runs as soon as the page has microphone access, before anything is being captured. Speak at the volume you plan to use and set the input gain so peaks land somewhere under the top of the meter.

Aim for peaks around -6 dBFS. Recording near full scale leaves nothing for a louder moment, and every stage after this one has gain of its own.

The input gain is applied before the file is written, so it is part of the recording rather than a preview. That is deliberate: a gain you can hear on the meter is the gain you get.

What the browser is not doing

The three processors a browser applies to a microphone by default are all switched off:

  • Echo cancellation, which removes anything resembling the speaker output.
  • Noise suppression, which gates and filters steady background sound.
  • Automatic gain control, which rides the level up and down on its own.

They exist to make voice calls intelligible. In a recording they are indistinguishable from damage: a gate that closes on a quiet phrase and an AGC that lifts room tone between sentences are both very hard to undo later.

Formats and depth

WAV writes uncompressed PCM at 8, 16, 24, or 32-bit. 24-bit is the sensible default for anything that will be edited afterwards: it costs half again as much disk as 16-bit and puts the quantisation floor far below anything a room will contribute.

MP3, M4A, Ogg, and FLAC are all encoded live as well. The lossy three are the ones that can fall behind on a slow machine, which is what the dropped blocks warning reports.

Long recordings

Audio goes to the encoder in blocks of about 170 ms and out to origin-private file storage as it is produced. Nothing accumulates in the tab, so an hour costs the same memory as a minute.

If the encoder does fall behind, blocks queue up to about eleven seconds and are then counted as dropped rather than held. A reported gap is recoverable information; a tab that runs out of memory is not.

Where the recording goes next

A finished recording hands off directly to the trimmer, the normalizer, or the audio workflow, carrying the file rather than making you find it again in a downloads folder. Trimming the dead air off both ends is usually the first thing worth doing.

Frequently Asked Questions

Those filters are built for voice calls. They gate quiet passages, duck one speaker under another, and reshape the top end, none of which belongs in a file you are going to measure or edit. The recording is what the microphone sent.

Yes. Audio is encoded as it arrives and written to origin-private file storage rather than held in the tab, so memory stays flat however long the recording runs.

The encoder could not keep up with the microphone and some audio was discarded rather than queued forever. It is rare, and almost always a lossy format on a slow machine. Record to WAV, or lower the sample rate.

Some devices only open at their own rate. When the browser refuses the requested one it falls back, and the rate the file was actually written at is shown next to the control.

Browsers hide device labels until microphone access has been granted at least once. Record once and the names appear.

Only on headphones. Monitoring through speakers puts the output back into the microphone, which builds into feedback within a second or two.

Explore Our Tools

Browse all tools