Convert WAV to Opus

Uncompressed CD-quality stereo costs 176,400 bytes a second — about 10 MB a minute, whatever is happening in the recording. Opus at 128 kbps costs 16,000 bytes a second, which is under 1 MB a minute and is a setting most people would call generous for music. Drop to 64 kbps for a stereo podcast and a minute is under half a megabyte.

Those numbers change what is practical rather than merely making files tidier. An hour-long interview is roughly 600 MB as a WAV and about 17 MB as Opus at the small-file band, which is the difference between a recording that lives on one machine and one you can attach to an email without thinking about it.

MP3 was standardised in 1993, AAC in 1997, Opus in 2012 as RFC 6716. Each generation had more computing power to spend at both ends and better models of what a listener actually notices, and Opus had the additional advantage of being designed with no obligation to remain compatible with anything.

The result shows up where the bitrate is tight. At 64 kbps an MP3 is audibly hollow and smeared; Opus at 64 kbps is a listenable stereo music file. The two converge as the bitrate rises, so at 256 kbps the argument becomes academic — which is why the case for Opus is strongest for exactly the people who came here to make files small.

Opus is unusual in having been built from two designs at once: a speech coder and a general-purpose transform coder, with the encoder choosing between and blending them according to what it hears. Older codecs had to pick one strategy and apply it to everything.

The practical effect is that you do not have to declare what your WAV contains. An interview with music under it, a lecture with a video clip played into the room, a podcast with a jingle — all encode sensibly at one setting. That is a genuine convenience when converting a folder whose contents you have not listened to.

Opus is defined around a 48 kHz sample rate. A WAV at 44,100 Hz — which is almost every WAV that came off a CD or out of a music application — is therefore resampled on the way into the encoder, and what comes out is a 48 kHz stream.

This is inaudible and it is worth knowing anyway, because it means an Opus file is not a candidate for a bit-exact comparison against its source and never can be. If a workflow downstream requires the original rate preserved exactly, the target you want is FLAC or WAV, both of which keep it.

A very large share of WAVs are stereo files carrying a single microphone duplicated across both channels: phone memos, conference recordings, single-source interviews, most text-to-speech output. There is no channel setting on this pair — the encoder is handed a codec and a bitrate and nothing else — so the Opus comes out with as many channels as the WAV went in with.

Downmixing to mono in an editor first would not help either, and that is the part worth knowing. The quality band fixes the bitrate from the codec alone, never from the channel count, so a mono source and a stereo source at the same band produce files of the same size. The band is the lever: the small-file setting is where an interview belongs, and on a large archive choosing it is worth more than anything else available on this page.

For audio delivered over the network, the size difference is the whole product decision: a page that ships 30 MB of audio and one that ships 3 MB behave differently on a phone on mobile data, and the second one gets listened to. Every current browser decodes Opus, so a single file usually covers the audience.

The exception is worth planning for rather than discovering. If your analytics show meaningful traffic from older devices, ship an MP3 alongside and let the markup choose; MP3 is the format nothing refuses. Serving one Opus file and hoping is a decision, and it should be a deliberate one.

Support is excellent in software and uneven in hardware. Car head units, older portable players, some televisions, some hi-fi streamers and older phones will show the file and decline to play it. The registry records Opus as having patchy support for precisely this reason, and it is not a problem that is going away quickly, because the affected devices are not being updated.

The test is quick and worth doing before a bulk conversion: convert one file, put it on the destination device, press play. If it plays, convert everything. If it does not, MP3 at 192 kbps is the fallback that will always work, at roughly three times the size.

The advantage of starting from a WAV is that the encoder sees the complete recording and every artefact in the result is introduced exactly once. That is the best position anyone converting to a lossy format can be in.

It follows that multiple outputs should all be made from this WAV rather than from each other. If you need an Opus for the web, an MP3 for the car and a small version for a transcript service, make three conversions from the source. Encoding an Opus from an MP3 compounds two sets of discarded audio, and the second encoder cannot tell which parts were missing already.

Each file is capped at 100 MB — a little under ten minutes of stereo WAV — and a drop takes up to a hundred files, returned together as a ZIP. Long recordings need splitting first; that ceiling is the browser holding the file in memory rather than an allowance anyone can raise.

Nothing is uploaded at any point, which matters here more than the arithmetic does. The WAVs people convert to Opus in bulk are interviews, therapy sessions, lectures, legal recordings and voice notes, and a conversion that keeps them on the machine they are already on is the only kind worth using for that material.