Which bitrate should I choose?
For speech — a lecture, an interview, a voice memo — 64 kbps is genuinely enough and 96 is generous. For music, 128 kbps is the usual floor and 192 is where most people stop hearing the difference on ordinary headphones. The tool shows what your file already is, so the useful question is not “what is good” but “how far below what I have can I go for this purpose”.
Why will it not re-encode at a higher bitrate?
Because it cannot put back what is not there. A lossy encoder threw detail away permanently the first time, and a second encoder cannot recover it — it can only spend more bytes describing the reduced version. You would get a file two or three times larger that sounds very slightly worse. Rather than take the instruction, the tool refuses it and shows the bitrate you already have.
Will it sound worse?
Yes, a little, and how much depends on how far you drop. Going from 320 to 192 kbps is inaudible to almost everyone; going from 128 to 64 is obvious on music and barely noticeable on speech. This is a second lossy pass over material that already lost something, which is why the honest use is a copy to send rather than a copy to keep.
What does the check before the download do?
It decodes the finished file again, from its own bytes, and requires three things: that it decodes at all, that the sample rate and channel count still match what went in, and that the length is within a frame of the original. If any of the three fails there is no download and the result says which — the way an audio file usually breaks is by playing as silence, and finding that out from the recipient is the expensive way.
Does my audio leave the device?
No. The decoding, the encoding and the check all happen in this tab. The encoder is a WebAssembly module served from this site — about 150 kilobytes, downloaded only when you open this tool — rather than a service somewhere that receives your recording.