CONCEPT Cited by 1 source

complexity tradeoff¶

Definition¶

Audio codec design is governed by a three-axis tradeoff: increasing quality generally requires either more bitrate (more bytes on the wire) or more complexity (more CPU per second of audio). "Good codecs achieve higher quality while balancing the other two" by exploiting signal structure and psychoacoustics (Source: sources/2024-06-13-meta-mlow-metas-low-bitrate-audio-codec, Figure 1).

The three axes¶

Quality — perceived fidelity of the decoded audio. Objectively measured via POLQA MOS on a 1–5 scale; subjectively via MOS listening tests.
Bitrate — compressed bits per second. Lower bitrate means more compression. In RTC, a bandwidth estimator drives this down as network quality degrades.
Complexity — encoder + decoder CPU/memory cost on the target device. Load-bearing for low-end device inclusion: a codec that only runs on flagships doesn't help the 20%+ of users on ARMv7 / 10+-year-old handsets.

Where MLow lands on all three¶

The 2024 Meta MLow announcement is unusually explicit that the codec wins on all three axes relative to Opus at the low-bitrate operating point:

Axis	Opus @ 6 kbps NB	MLow @ 6 kbps WB
Quality (POLQA MOS)	1.89	3.9
Bitrate	6 kbps	6 kbps
Complexity	baseline	~10% lower

This is the "achieve higher quality while balancing the other two" ideal, delivered concretely.

ML codecs and the constraint¶

ML-based neural codecs like Meta's Encodec relax the bitrate / quality tradeoff dramatically but at the cost of complexity. Meta's MLow decision is a clean case study of the constraint biting back: "only the very high-end (expensive) mobile handsets are able to run these codecs reliably" means the complexity axis collapsed the deployable user population. See patterns/classic-dsp-over-ml-for-compute-constrained.

Seen in¶

sources/2024-06-13-meta-mlow-metas-low-bitrate-audio-codec — explicit Figure 1 framing of the triangle and canonical worked example of a codec that wins on all three axes.