Why an online voice recorder should not upload anything
Almost every site offering this records to their server. That is an odd default for the file type in question. A voice recording identifies you, carries whatever you happened to say, and increasingly is enough material to synthesise your voice.
There is no technical need for it. MediaRecorder has been in every major browser
for years and produces a finished, playable file locally. So this records into the
page, plays it back from the page, and hands it to you as a download. The only
copy that exists is the one in your tab, until you save it.
The practical consequence: reloading loses the recording. That is the trade for there being no server copy, and it is the right way round.
Getting the level right in a browser voice recorder
The meter is there because level is the one thing you cannot fix afterwards.
Aim for peaks between −6 and −12 dBFS. Loud enough to sit well clear of the room noise, quiet enough that a laugh or an emphasised word has somewhere to go.
If the top bars light red, you are clipping. The waveform has hit the ceiling and the peaks are flat-topped: that information is gone and no amount of editing brings it back. Turning it down later gives you a quiet distorted recording. Move back from the microphone or reduce the input gain and record it again.
What the meter shows when you stop talking is your noise floor. A low steady reading is normal. If it lights several bars, something in the room is loud or your input gain is too high.
Recording something worth using
Three things account for most of the difference between a usable recording and a poor one, and none of them is equipment.
Distance. A hand's width from the microphone, slightly off-axis so plosives do not hit it straight on. Further away and you record more room than voice.
The room. Soft surfaces absorb reflections; hard parallel walls create them. A room with a rug, curtains and furniture sounds dramatically better than a tiled bathroom or an empty office, with the same microphone.
Continuity. If the recording is a reference for a voice model or a narration bed, record continuous natural speech rather than isolated phrases, and leave a couple of seconds of silence at the start so whatever processes it later has a sample of your room tone.
What the format means
The tool picks the best container your browser supports and names the file to match. WebM with Opus is what Chrome and Firefox produce; MP4 with AAC is what Safari produces. Both open in every editor worth using, and both are far more efficient than WAV for speech.
If you specifically need WAV, record here and convert afterwards. Recording uncompressed in a browser tab is possible but consumes memory fast for no audible benefit on speech.