Skip to content

SYSTEM Cited by 1 source

Meta Encodec

Definition

Encodec is Meta's open-source ML-based audio compression codec, released October 2022 by Meta FAIR. It uses a learned neural-network encoder/decoder to compress audio at very low bitrates with "amazingly crisp audio quality" relative to classical codecs (Source: sources/2024-06-13-meta-mlow-metas-low-bitrate-audio-codec citing the Meta AI blog post).

Why it's on the wiki

Encodec is the canonical ML-codec example Meta's RTC team explicitly chose not to deploy for real-time calling. The 2024-06-13 MLow announcement names Encodec as the exemplar of the class and then rejects the class:

"While these AI/ML-based codecs are able to achieve great quality at low bitrates, it often comes at the expense of heavy computational cost. Consequently, only the very high-end (expensive) mobile handsets are able to run these codecs reliably, while users running on lower-end devices continue to experience audio quality issues in low-bitrate conditions."

This is a load-bearing counterexample: Meta built a second, non-ML codec (MLow) rather than ship its own ML codec into WhatsApp / Instagram / Messenger because of low-end device inclusion. See patterns/classic-dsp-over-ml-for-compute-constrained.

Trade-off named in the MLow source

  • Quality at very low bitrates: ML codecs (Encodec-class) are strong.
  • Compute on target devices: ML codecs require flagship handsets; Meta has >20% of calls on ARMv7 and 10s of millions of daily WhatsApp calls on 10+-year-old phones.
  • Net user impact: "limited to a small portion of users" — therefore not the right deployment choice for billions-of-users RTC.

Not covered here

Encodec's architecture, quantization scheme, or training setup are not described in the MLow source. This page is a reference stub for the codec-class contrast that motivated MLow, not a deep treatment of Encodec itself.

Seen in

Last updated · 319 distilled / 1,201 read