A WAV file stores sound uncompressed: every sample is written out in full, silence included. At CD quality, stereo at 44.1 kHz and 16 bits, that comes to about 10 MB for each minute, so an hour-long recording weighs around 600 MB. Recorders set to 48 kHz or 24 bits produce even more.
MP3 removes the parts of the signal the ear is least likely to notice and stores the rest compactly. The same minute takes roughly 1 MB at 128 kbps, 1.4 MB at 192 kbps and 2.4 MB at 320 kbps. Against a CD-quality WAV that is a reduction of about eleven, seven and four and a half times. The often quoted "ten times smaller" is true at 128 kbps and optimistic at the higher settings.
For speech, such as interviews, lectures, dictation and meetings, 128 kbps is entirely sufficient. A voice occupies a narrow part of the spectrum and nothing useful is gained above that.
192 kbps is the default because it is a safe middle: most listeners cannot tell it from the original on ordinary headphones or speakers, with speech or with music. 320 kbps is the highest bitrate MP3 offers. It is the right choice for music you will listen to closely or hand to someone else, at the cost of a file more than twice the size of the 128 kbps version. None of the settings improve the recording; they only decide how much of it is kept.
MP3 compression cannot be undone. Converting the MP3 back to WAV gives a large file with the reduced quality, not the original. So treat the MP3 as a copy for sharing and listening, and keep the WAV in three situations: when you will edit, mix or master the audio further, because every new lossy export degrades it a little more; when it is an archive or master recording; and when a client, a broadcaster or a distributor asks for uncompressed delivery.
If the only issue is storage and you need no loss at all, a lossless format such as FLAC roughly halves the size. This tool does not produce FLAC; its output is always MP3.
Uploading a 600 MB WAV to an online converter can take longer than the conversion itself. Here there is no upload. The page runs ffmpeg, compiled to WebAssembly, inside the browser: the WAV is read from your disk, encoded in memory and saved as an MP3. The engine, about 30 MB, is downloaded once on first use and cached.
Encoding speed depends on the device and on a laptop is normally many times faster than real time. Memory is the practical limit: ordinary recordings are no problem anywhere, while WAV files of several gigabytes, such as multi-hour sessions, may fail on a phone and are better converted on a computer.
Many large WAV files are spoken recordings waiting to be written up. Once the MP3 exists, a "Transcribe this file" button sends it to Transcrify's free transcription. That step does upload the file, to Transcrify's own servers, where the speech recognition runs; this is where the smaller size pays off, since the MP3 uploads in a fraction of the time the WAV would need.
10 recordings are free, each up to 3 hours and 5 hours in total, shared between guest use and a free account. A guest receives a timestamped preview, the full text of short recordings and around the first ten minutes of long ones; a free account keeps the full transcript.