Convert MP4 to Opus

Opus was published as RFC 6716 in 2012 and designed for real-time communication, which meant surviving at bitrates nobody had previously considered usable. Every browser implements it because every browser needed it for calls, and that pedigree is exactly what makes it good at the material people pull out of video.

Most extracted audio is somebody talking: a lecture, a sermon, an interview, a webinar, a tutorial. On that material Opus stays clear and natural at sizes where MP3 turns watery and metallic. For a single podcast episode the difference is a curiosity; across a term of recordings on a phone it is the difference between fitting and not.

The picture is the expensive part of a video and it is discarded here, so the result is governed almost entirely by the length of the recording and the quality band. A two-hour lecture that arrived as a two-gigabyte MP4 produces an audio file measured in tens of megabytes, and at the small band a good deal less than that.

The three bands move the figure by roughly a factor of two in each direction. For speech, the small band is genuinely usable rather than a compromise — this is the codec designed to work there. For music, start at balanced and go up, because low-bitrate performance on dense material is not what Opus was optimising for.

Both are Opus audio inside an Ogg container, written by the same encoder on this site. Nothing about the bytes differs in a way a decoder cares about. What differs is what software assumes when it sees the name.

Take .opus when the codec was your decision — a voice archive, an app’s audio assets, a corpus you are building. Take .ogg when a piece of software, a game engine or an upload form asked for that extension by name, because some of those will refuse an unfamiliar suffix even for a file they can decode perfectly.

Every current browser plays it, which makes it the natural choice for audio served from a web page. Android handles it natively, VLC and Audacity read it, and the messaging apps that record voice notes have been using it quietly for years.

Apple came to it later and less completely, and appliances have not come to it at all: car stereos, hi-fi separates, older portable players and gym equipment will show an unplayable file. That is the trade for the size, and it is a fair one when the destination is a phone or a computer and a bad one when the destination is a device with a physical volume knob.

The audio in the MP4 is already lossy AAC. Encoding it as Opus decodes that and compresses again, so a second helping of detail goes. There is no way around it short of not converting, and the honest thing is to say so before the click rather than after.

On speech at the balanced band it is inaudible by any test you are likely to run, because both encoders make similar decisions about what a voice does not need. On applause, cymbals and dense mixes at the small band it is audible, and the fix is a higher band rather than a different format.

Up to a hundred files at a time on the free tier, each converting independently, with the results downloadable together as a ZIP. Nothing waits behind anybody else because there is no shared server to queue on.

This is where the size argument compounds. A language-learning deck, an oral-history collection or a set of voice assets is hundreds of short clips, and a format that is a third of the size of the obvious alternative changes what you can carry, sync and back up rather than saving a rounding error on one file.

A lecture recorded on a single microphone is usually mono information delivered as two identical channels, and encoders handle that efficiently but not for free. If your source is genuinely single-microphone, an editor that can fold it to mono before compression is worth the step on a large collection.

The other thing worth a listen before committing to a batch is the first minute of one file at the band you intend to use. Recordings with a lot of room noise, air conditioning or overlapping speakers are harder to compress than a clean voice, and the band that sounded fine on a studio podcast can sound thin on a phone recording of a seminar.

A video can carry more than one audio track — an original language and a dub, a commentary, separate microphone feeds — and every one the browser can decode is written into the Opus file, in the order the MP4 held them. A track in AC-3, E-AC-3, DTS or TrueHD is dropped without a warning. For anything recorded on a phone, a screen recorder or an ordinary export there is only one and the question never comes up.

For a professionally delivered file it is worth checking what is inside before building a collection on the result. A lecture capture holding both the room microphone and a lapel produces a file holding both, and players differ in how readily they let you switch between them; a batch of two hundred built on the wrong assumption is discovered late and redone entirely.

The decode and encode happen inside the browser tab on your own processor, so nothing carrying the video is sent anywhere. There is no account, no daily count, and the free tier takes files up to 100 MB each.

For a collection of interviews or fieldwork recordings that is the point rather than a bonus. Consent for a recording rarely extends to a converter site nobody was told about, and doing the work locally means the question does not arise.