Convert OPUS to MP3

Opus was published as RFC 6716 in 2012 by Xiph.Org and is the mandatory audio codec for WebRTC, which is the reason it is everywhere without being famous. Every browser had to implement it to support calls, and once every browser had it, every app built on browser technology used it: WhatsApp and Telegram voice messages, Discord voice and its recordings, Signal, Google Meet, in-browser recorders, and `yt-dlp` whenever it picks the best available audio stream.

That is why this page has a different reader from the rest of the audio bundle. Someone converting an M4A knows they have an Apple file. Someone converting `.opus` usually found the extension for the first time when they copied something off a phone or out of a downloads folder, and their question is not about codecs — it is "why will nothing open this".

Here is the number worth carrying away. A voice note is mono and Opus encodes speech at somewhere between 16 and 32 kbps while remaining perfectly clear; that is the whole reason messaging apps chose it. A two-minute voice note is therefore around 300 kilobytes.

Re-encode that at High quality and the MP3 is written at 320 kbps: roughly 4.8 megabytes, thirteen to twenty times the original, containing precisely the same amount of information. The extra bits describe nothing, because there is nothing left to describe — the detail was never captured. Small file at 96 kbps is the correct setting for speech and is still three to six times the source bitrate. This is the one page in the bundle where the highest setting is actively the wrong answer.

Two lossy encodes in a row always cost something, and this pairing costs a little more than most because of the direction. MP3 dates from 1993 and Opus from 2012, and nineteen years of research sit between them — at 64 kbps Opus is doing work MP3 cannot do at all. So the second encoder is not merely re-compressing; it is a less capable tool re-compressing.

For speech, which is what most of these files are, the result is fully intelligible and marginally duller — a slight loss of air on sibilants and a flatter room sound. For music captured at a low browser bitrate, the degradation is more noticeable than the equivalent step from AAC would be. Neither is a reason not to convert when something downstream requires MP3. Both are a reason to keep the `.opus` original.

Transcription is the most common destination. A good many services, Whisper wrappers and forced-alignment tools accept a short list of formats and Opus is often not on it, not because the audio is unsuitable but because their ingest code predates the format. MP3 or WAV clears the requirement, and for transcription MP3 at 96 kbps is entirely adequate — the model is reading words, not judging fidelity.

The other destinations are editing and delivery: a DAW or podcast editor that refuses the file, a client or colleague who needs to double-click it and have it play, a hardware recorder or car system, an upload form with a fixed list. In every one of those the requirement is about the receiving software, which is worth knowing because it tells you the conversion has done its job the moment the file opens.

Voice notes and call recordings are almost always a single channel, and the conversion keeps it that way. A mono MP3 is half the size of a stereo one at the same bitrate, and there is no benefit whatsoever to duplicating one channel into two — it doubles the file and adds nothing.

If some receiving program insists on stereo, let it do the duplication on import, where it costs nothing on disk. Opus itself supports up to 255 channels against MP3's two, so at the other extreme a multichannel Opus capture is folded down to stereo at 48,000 Hz rather than refused. That case is rare outside broadcast experiments, but it is what happens.

Opus in an Ogg stream carries Vorbis-style comment tags, and where they exist they are mapped onto ID3v2 in the MP3 — title, artist, album, track number, date, comment — along with any embedded picture, which becomes an APIC frame.

In practice most of these files arrive with nothing in them. A voice note is named by the app with a timestamp and carries no metadata at all, so the MP3 comes out untagged and that is faithful rather than lossy. If you are archiving a series of recordings, the filename is doing the work the tags are not, and it is worth naming them before conversion rather than after.

Ogg Opus records a pre-skip value so that a player knows exactly how many samples at the front are encoder priming and should be discarded, which is how a stream can start cleanly. MP3 has no equivalent that anything reliably reads.

The MP3 written here carries a Xing header with the frame count, the file size and a seek table — enough for accurate seeking and a correct duration — but not the delay and padding fields that would let a player trim the encoder's own padding. A few dozen milliseconds of silence are therefore added at each end. Irrelevant for a voice note; relevant if you are aligning a converted recording against video or against a second track, where the fix is to align in the editor after conversion rather than to trust the file.

This is the pair where the argument stops being abstract. The files are voice messages, recorded calls, therapy notes, interviews, meetings that were captured with permission for one purpose. Handing them to a conversion site means handing a stranger a copy, and the terms of service you did not read decide what happens next.

Here the decode uses the Opus decoder your browser already contains — it is the same one that renders a call — and the MP3 is produced by a WebAssembly build of the LAME encoder loaded into the tab, because no browser can encode MP3 by itself. Neither step needs a server, so there is no upload, no queue, no account and no record anywhere of what you converted. The claim is checkable in thirty seconds with the network tab open, which is the only kind of privacy claim worth making.

Phone backups produce these by the hundred. Drop a hundred of them at a time: each file converts on its own and they come back as a single ZIP, with a 100 MB ceiling per file — far above anything a voice note approaches — and a hundred files to a batch, which is the limit that actually bites on a folder of these.

Keep Small file selected for a batch of this kind. Four hundred voice notes, four batches of them, at 96 kbps is a few hundred megabytes; the same four hundred at 320 kbps is well over a gigabyte of MP3 carrying no additional information whatsoever. This is the setting decision with the largest practical consequence anywhere in this bundle.