Skip to content

Add section: AI in encoding — per-title encoding and VMAF optimization #17

Description

@alexbj75

Summary

The existing "AI and Machine Learning" section contains only two links, both from 2018, focused on video AI platforms and metadata. It completely misses the 2024–2026 story: AI-driven encoding is now production-mature and is changing how streaming services deliver video quality.

What's missing

Per-title (and per-scene) encoding

Netflix's per-title encoding approach — using ML to analyze content complexity and generate an optimal ABR ladder per title — is now baseline for premium OTT. The concept needs a clear explanation:

  • What it is (vs. fixed ladder encoding)
  • Why it matters (40–50% bitrate savings at same quality, or higher quality at same bitrate)
  • How AI accelerates it in practice

VMAF as quality measurement standard

VMAF (Video Multi-Method Assessment Fusion) developed by Netflix is now the standard loss-quality metric for encoding optimization. Any streaming engineer working with encoding quality needs to know it exists and what it measures.

AI-accelerated AV1 encoding

Netflix and Meta report 40% faster AV1 encoding with ML-optimized pipelines. This is directly connected to AV1's adoption trajectory (see also related issue on AV1 positioning).

AI subtitling

AI subtitling now covers 50–60% of live broadcasts at leading European broadcasters (SVT, NRK). Whisper and commercial alternatives (AWS Transcribe, Google Speech-to-Text) are production-ready. The media supply chain workflow (transcode + package + subtitle) is a standard automation pattern.

Eyevinn's auto-subtitles tool is a directly relevant OSS reference: https://github.com/Eyevinn/auto-subtitles

Suggested approach

Replace or substantially expand the "AI and Machine Learning" section with 2024–2026 relevant content covering:

  • Per-title/per-scene encoding and VMAF
  • AI subtitling (with Eyevinn auto-subtitles as reference)
  • AI-driven media supply chain orchestration

References

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

enhancementNew feature or requesthelp wantedExtra attention is needed

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions