Tutorials Developer

Jean-Marc Valin: One Codec for Speech and Music

By Infinite Creation · Profile

At a Glance

Before Opus, an application had to choose. Speech codecs were efficient on voice and poor on music. Music codecs were the reverse, and both introduced delay that made conversation awkward. Opus, published as RFC 6716 in September 2012, does both in one format and switches between the two techniques as the audio changes. It is now the default for WebRTC, and therefore for most voice and video calls made in a browser.

Lived: born 1975, Canadian
Anchor year: 2012, Opus published as RFC 6716
Strand: overlooked

Career

The problem as it stood

Two families of codec had evolved separately. Linear prediction suits speech, because the human vocal tract is a filter that can be modelled cheaply. Transform coding suits music, because it can throw away what the ear will not hear. Neither did the other job well, and switching formats mid-call was not practical.

What he actually did

Valin's CELT work on low-delay transform coding was combined with Koen Vos's SILK speech codec into a single format with a hybrid mode. The result covers six kilobits per second for narrowband speech up to 510 kilobits for stereo music, and the encoder can move between modes without the listener hearing a seam.

How it worked

The encoder decides, frame by frame, whether the signal is better served by linear prediction, by the transform, or by both at once. Low delay was treated as a first-class requirement rather than an afterthought, which is what makes it usable for conversation rather than only for playback.

What it made possible

Opus is mandatory to implement in WebRTC, so every browser-based call carries it. It is royalty free and specified in an open standards document, which is the reason it spread into games, streaming and broadcast contribution links rather than staying in one vendor's product.

Where the credit landed

RFC 6716 names three authors: Valin and Timothy B. Terriberry of Mozilla, and Koen Vos of Skype. Popular accounts often name one of them. The document names three, and the format is the join of two separately developed codecs rather than one person's design.

Known For

Opus is mandatory to implement in WebRTC, which is why it is everywhere without being marketed.
The hybrid mode is the interesting part: it is one format doing two jobs, not two formats bundled.
It is royalty free, and that is a design decision rather than an accident.

Knowledge Base

What You'll Learn

Opus is where this corpus's speech strand and its music strand finally stop being separate.

Two lineages, one format

The linear-prediction half descends from decades of telephony research into modelling the vocal tract. The transform half descends from Ahmed's discrete cosine transform and the psychoacoustic work of Fletcher and Schroeder that made perceptual coding possible. Opus is the point where a single encoder decides, frame by frame, which lineage to use.

Where This Fits

Opus compresses speech and music for delivery, a modern transmission link in the chain covered in A History of Broadcast Technology.

FAQ

Q: Is Opus better than MP3?
A: For its purposes, substantially. It was designed twenty years later for interactive use, handles low delay and packet loss, and is royalty free. MP3 remains more widely supported in old hardware, which is a different kind of advantage.

Q: Why is Opus in every browser?
A: Because WebRTC makes it mandatory to implement. A specification that requires a codec guarantees interoperability in a way that recommending one never does.

Vocabulary

Opus An audio codec published as RFC 6716 in 2012, covering everything from narrowband speech at six kilobits per second to stereo music at 510. It is mandatory to implement in WebRTC, which is why almost every browser call uses it.

Linear prediction A coding technique that models the human vocal tract as a filter and transmits the filter settings rather than the waveform. Very efficient on speech and poor on music, which is why speech and music codecs developed separately.

CELT The low-delay transform codec Valin developed, which became the music-handling half of Opus. Its contribution was treating latency as a design requirement rather than something to be traded away for efficiency.

Royalty free Usable without paying a patent licence. For a codec this determines adoption more than quality does, because it decides whether a browser or an open-source project can ship it at all.

Sources

Every record below was fetched and read before it was cited here. Checked 26 August 2026.

RFC 6716, Definition of the Opus Audio Codec. Standards Track, published September 2012. The header names three authors: Jean-Marc Valin and Timothy B. Terriberry of Mozilla, and Koen Vos of Skype Technologies. The abstract gives the range as six kilobits per second for narrowband speech to 510 kilobits for stereo music, and states that the codec uses both linear prediction and the modified discrete cosine transform.
Y. Chen and others, An Overview of Coding Tools in AV1. APSIPA Transactions on Signal and Information Processing, 2020. Valin is one of twenty-one named authors, which is worth seeing: modern codec work is committee work, and the author list is the evidence.