Convert WAV to MP3
Shrink a recording from tens of megabytes to a few.
How it works
A WAV file stores sound the way a microphone measured it: a stream of numbers, one per channel, taken tens of thousands of times a second. Nothing is compressed. CD audio takes 44,100 measurements a second at 16 bits across two channels, which works out at 1,411,200 bits every second, or about 10.1 MB per minute. An hour-long recording is 600 MB and a podcast episode will not fit in an email.
That arithmetic is the whole reason this conversion exists. The samples themselves are already as good as the recording gets, so the question is never quality against quality. It is how much of the file can be thrown away before a listener notices.
What the encoder actually removes
MP3 works on the limits of hearing rather than on the waveform. Loud sounds mask quieter ones nearby in frequency and in time, so a cymbal crash hides the room tone underneath it and a listener cannot report what was never audible. The encoder models that masking, splits the signal into frequency bands, and spends its bits only where the ear could detect their absence.
It also exploits the two channels being nearly identical. Joint stereo encodes the sum and the difference rather than left and right separately, and the difference signal is small and cheap. On a mono source recorded to a stereo file, which is most voice recordings, this saves close to half the bits for nothing.
Choosing a bitrate honestly
128 kbps was the format's default for a decade and gives roughly 11:1 compression against CD audio. Most listeners on most equipment cannot pick it out of a lineup with the original for pop or spoken word. Trained ears on good headphones can, on cymbals and applause, where the masking model has least to work with.
192 kbps removes nearly all of that doubt for music. 320 kbps is the format's ceiling and is worth choosing when the file is an archive copy rather than a delivery copy. For speech, 96 kbps in mono is generous, and drops an hour of talking to about 43 MB.
One thing to get right first: this is the last step. Editing, mixing and noise removal all belong on the WAV, because every operation performed on an MP3 requires decoding and re-encoding it, and the loss compounds each time. Keep the WAV and treat the MP3 as an export.
Questions
Is my recording uploaded?
No. The LAME encoder is compiled to WebAssembly and runs in a Web Worker on your own machine. Open the Network tab in developer tools before you convert: you will see the page's own code and the encoder module download once, and no request afterwards. There is no server here that could receive a file.
Is there a file size limit?
No practical one. WAV is read as a stream rather than held in memory all at once, so a multi-gigabyte session file works, and encoding runs at roughly fifty times real time. A ninety-minute recording finishes in a couple of minutes on a laptop. The one real constraint is that the finished MP3 has to fit in memory before it is handed to you, and MP3s are small by definition.
My WAV is 24-bit or 96 kHz. Does that matter?
Not to the MP3. The encoder works on the decoded signal and writes its own representation, so bit depth disappears entirely and any sample rate above 48 kHz is resampled down, since MP3 does not define rates higher than that. High-resolution masters convert fine; what makes them high resolution does not survive, which is expected and is the point of the format.
Why is my MP3 slightly longer than the WAV?
MP3 encodes in fixed blocks of 1,152 samples and the encoder's filter bank introduces a delay, so the last block is padded with silence and a short lead-in appears at the start. The encoder writes a header recording exactly how much padding it added, and players that read it trim both ends back to the original length. Older hardware players that ignore that header will show a fraction of a second more.