Convert WebM to WAV

This page has a narrower reader than most. A WAV is not something you put on a phone or send to a friend — it is four to twenty times larger than the file it came from and sounds no different. It exists because some piece of software on the other side of the conversion has told you, explicitly or by refusing everything else, that it wants uncompressed PCM.

That software is usually one of four things: a transcription or speech-recognition model, an audio editor like Audacity or a DAW, an analysis script built on a library that reads WAV and nothing else, or a telephony or accessibility system with a documented input format. All of them are doing further work on the samples, and all of them would rather start from something that has not been compressed a second time.

16-bit PCM in a RIFF WAVE container — the format Microsoft published in 1991 and the one every audio tool on earth reads without argument. The sample rate and channel count are whatever the WebM already had, carried across unchanged rather than resampled. Browser-recorded Opus runs at 48 kHz, so a voice note typically arrives as 48 kHz mono and a captured meeting as 48 kHz stereo.

Sixteen bits is the depth that matches what came in. The source was a lossy stream decoded to floating point; writing 24 or 32 bits would store the decoder’s rounding error at greater precision and produce a file half again as large with nothing extra in it. If your pipeline requires a specific depth or rate, resample in the tool that needs it, where the resampler is one you can inspect.

Every other audio target on this site offers a quality band, because a lossy encoder has to be told how many bits per second it may spend and the right answer differs between codecs by more than a factor of two. WAV has no such dial, because PCM is not a codec. It is the samples, written down.

The only two things that could be changed are the sample rate and the bit depth, and changing either is resampling rather than converting — a different operation with its own quality trade-offs, which belongs in the tool that has a reason to do it. So the control simply is not offered, and that absence is the honest answer rather than an oversight.

This is the misconception the page exists to correct. Converting a compressed recording to a lossless format does not restore anything. Opus threw away parts of the signal it judged inaudible, permanently, before the file ever reached you. The WAV preserves what survived with perfect fidelity and preserves nothing else.

The value is forward-looking rather than backward. From the WAV onward, every edit, every filter and every export starts from the same samples with no further generational loss. That is exactly what an editor or a transcription model wants: not a better recording, but a guarantee that the tool is not fighting compression artefacts that a second lossy pass would have added.

Uncompressed audio costs a constant number of bytes per second: 96 kB per channel per second at 48 kHz and 16 bits. That works out at roughly 5.8 MB a minute in mono and 11.5 MB a minute in stereo, and it does not vary with how loud, complex or quiet the recording is. A one-hour interview in mono is about 350 MB.

Compare that with the WebM it came from, which for the same hour of speech might be forty megabytes. The ratio is startling the first time you see it and it is not a bug in either format — Opus is extremely good at speech, and PCM makes no attempt to be good at anything except being exact. Have somewhere to put the file before you convert an afternoon of recordings.

Speech-recognition tools are the most common destination for this conversion, and most of them accept WAV first and everything else grudgingly. Whisper and its many wrappers, commercial transcription services, and the audio libraries underneath both, all document PCM as the input they expect and convert anything else themselves before they start.

Doing that conversion yourself is worth it for one reason: you get to see the result. If the WAV turns out to be silent, half a second long, or the wrong one of two microphone tracks, you find out in an audio editor in ten seconds rather than in a transcript full of nonsense an hour later. Converting is also where you discover that a recording has no audio track at all, which happens more often than anyone expects.

A WebM can hold several audio tracks and a WAV file, as written here, holds one. The first track in the container is the one that survives. Meeting recorders that write the microphone and the shared system audio as separate tracks are where this bites, because you get one of them and no warning about the other.

If the recording matters enough to transcribe, open the WAV before you rely on it. Where both tracks are needed, Audacity reads the WebM container directly with its FFmpeg component installed and can import the tracks side by side — the extra install is worth it once for a recording you cannot make again.

A WAV of an hour-long call is hundreds of megabytes, and uploading that to a converter and pulling it back is the slowest part of the whole job by a wide margin. Here the decode and the write both happen in the tab, at the speed of your own disk, so the length of the recording is the only thing that costs time.

It also removes the awkward question that hangs over this pair in particular. Recordings destined for transcription are calls, consultations, interviews and depositions — the exact material that should not be handed to an unfamiliar service. Nothing carrying the file is sent anywhere, and the network tab will show you that in the time it takes to convert one.