Now Streaming LOCMAF
Low Overhead CMAF for MOQT

✨ Why LOCMAF

CMAF carries the metadata that DRM needs (per-sample IVs, subsample maps, tenc defaults). LOC is compact but strips this away. LOCMAF keeps the CMAF structure intact and compresses the redundant moof bytes out of the wire — so the standard MSE / EME path on the receiver plays protected content as if it had arrived as plain CMAF.

• Sample-level objects (one frame per MoQ object)
• Delta moof as small as 2 bytes
• Functionally lossless reconstruction
• Works with Widevine, PlayReady, FairPlay, ClearKey

📊 Measured Compression

On a 250-fragment AVC group from the moqlivemock test asset:

• CMAF moof total: 26 040 B
• LOCMAF object total: 592 B
• moof-header ratio: 45 : 1
• 128 kbps AAC wire bitrate: 171.5 → 131.9 kbps (−23 %)
• 128 kbps AAC under cbcs DRM: 191.4 → 131.9 kbps (−31 %)

LOCMAF saves more relative to CMAF under DRM than in clear, because senc / saio / saiz make the CMAF moof grow while LOCMAF only emits what it needs.

See it yourself: after connecting MOQ Player to the server, click the Show catalog toggle to inspect each track's total wire bitrate (framing included) and compare a CMAF track to its LOCMAF counterpart — both renditions live in the same mlm/cmsf/* catalog.

🔗 LOCMAF Resources

Everything you need to understand and implement LOCMAF:

• locmaf.dev

  overview, diagrams, measurements
• draft-einarsson-moq-locmaf

  IETF Internet-Draft — normative LOCMAF spec (wire version 0.3)
• Eyevinn/locmaf

  Go reference codec (encode/decode + golden vectors)
• Slide deck

  20-slide MARP deck adapted from the spec
• Eyevinn/moqlivemock

  MoQ test publisher/subscriber (uses the locmaf codec)
• Eyevinn/warp-player

  browser MoQ player with LOCMAF + EME

Catalog signal: "packaging": "locmaf" with "locmafVersion": "0.3" in the CMSF Track entry. The matching "packaging": "cmaf" rendition shares the same init data through an initRef, so both variants sit side by side in one draft-01 catalog.

📝 What is Published

The publisher generates WebVTT (wvtt) and TTML (stpp) subtitle tracks that show the UTC time and group number. A subtitle group holds one CMAF chunk per video object — 25 a second for the test content — and each chunk is sent when the interval it covers ends, so text reaches the player with the latency of the picture it belongs to.

• In every mlm/cmsf/* namespace — clear, DRM and ClearKey; the LOC (mlm/msf/clear) and moq-mi namespaces carry video and audio only
• Every subtitle track also as LOCMAF, <name>_locmaf, sharing the CMAF track's init data
• A cue keeps its true begin, and gets its end only in the chunk where it ends, so an unchanged cue restates byte for byte
• Each track's measured bitrate in the catalog, and each format in its own altGroup

🎨 Paint-Model Subtitles

At 25 chunks a second, plain stpp restates a complete TTML document 25 times a second. The experimental paint-model variants stpc and wvtc send a chunk that changes nothing as an 8-byte ttmn or vttn box, and a changed TTML chunk as its <body> alone in a ttmb box. Wire bitrate, CMAF / LOCMAF:

• stpp 317 / 296 kbit/s  →  stpc 38 / 17 kbit/s
• wvtt 35 / 14 kbit/s  →  wvtc 25 / 4.5 kbit/s

This demo publishes them as subs_stpc_en and subs_wvtc_sv. mlmpub leaves them off by default, since the 4CCs are unregistered placeholders: -subsstpc en -subswvtc sv adds them.

▶️ Rendering in MOQ Player

MOQ Player reads each subtitle object itself, CMAF or LOCMAF, and paints the cues on the same overlay as the CTA-608 captions, in step with the picture. TTML is laid out by imscJS, the IMSC renderer dash.js uses; WebVTT by the player itself.

• One Subtitles / CC selector: Off, CC1 or one subtitle track at a time
• The paint-model formats are recognised by the sample entry of the init segment, and a no-change sample is shown again without parsing
• Measure all subtitle tracks side by side shows, per track, the bitrate and what the player parses per second

💬 What is Carried

The publisher injects one self-contained CTA-608 CC1 caption per MoQ group, generated with Eyevinn/go-608. Row 13 carries the UTC clock, row 14 the group number, so what you read on screen is directly checkable against the burnt-in overlay.

• AVC and HEVC — an SEI NAL unit before the first VCL NALU
• AV1 — a metadata_itu_t_t35 OBU before the first frame OBU
• Present in CMAF, LOCMAF, LOC and moq-mi
• Spliced in before encryption, so captions ride inside the ciphertext
• Each group is independent — a caption comes from that group's samples alone

🎛️ Presentation Modes

mlmpub -cc608mode selects how a caption reaches the screen, covering all three CTA-608 presentation styles. Measured at 25 fps in a 25-frame group:

• paint-on (default) — grows two characters per frame; first characters on frame 4, complete on frame 17, so the caption is displayed over the second it names
• roll-up2 / roll-up3 / roll-up4 — types the same way but scrolls the window up before each line, finishing two frames later
• pop-on — builds invisibly and flips on whole at frame 18

The roll-up sizes differ in the RU2/RU3/RU4 mode code on the wire but display identically here, since a group holds one cue and the window is reset per group — which is exactly what makes them useful for exercising a decoder.

▶️ Rendering in MOQ Player

MOQ Player decodes the captions from the in-band SEI on both render engines — MSE (CMAF and LOCMAF) and WebCodecs (LOC) — and paints them on the true 32×15 grid inside the CTA-608 safe area, with colours, background boxes and the full pop-on / roll-up / paint-on screen model.

• CC1 is one choice in the Subtitles / CC selector, offered when the track advertises urn:scte:dash:cc:cea-608:2015 and its codec can carry CTA-608; the CC button is a shortcut
• Encrypted namespaces work without a key — the caption SEI sits in the clear subsample leader
• Built on a general timed-text overlay seam, which now also paints the WebVTT and IMSC subtitles

Not yet on AV1: AV1 renditions play on the WebCodecs pipeline, but their captions travel in a metadata OBU rather than an SEI NAL unit, so the player does not render them — CC1 is greyed out in the selector on those tracks, with the reason.

🎥 Video Tracks

H.264 (AVC), HEVC and AV1 encoded video streams at multiple bitrates for adaptive streaming, each carrying in-band CTA-608 captions

• 400 kbps AVC
• 600 kbps AVC
• 900 kbps AVC
• 400 kbps HEVC
• 600 kbps HEVC
• 900 kbps HEVC
• 400 kbps AV1
• 600 kbps AV1
• 900 kbps AV1

All three codecs are served in the mlm/cmsf/* namespaces (CMAF and LOCMAF) and in LOC; mlm/moq-mi/clear video is AVC only. The AV1 assets are encoded in low-delay CBR mode (I/P only, no reordering) so they match the AVC/HEVC structure.

🎵 Audio Tracks

Easily discernable audio tracks in AAC, Opus, and AC-3 with synchronized second beeps

• Monotonic AAC (128 kbps)
• Scale AAC (128 kbps)
• Monotonic Opus (128 kbps)
• Scale Opus (128 kbps)
• Monotonic AC-3 (192 kbps)
• Scale AC-3 (192 kbps)

📝 Text Tracks

Out-of-band subtitle tracks in the mlm/cmsf/* namespaces, one object per video frame in CMAF and LOCMAF, plus captions carried inside the video itself

• WVTT (WebVTT in CMAF), subs_wvtt_sv
• STPP (TTML in CMAF), subs_stpp_en
• WVTC and STPC, the experimental paint-model variants, subs_wvtc_sv and subs_stpc_en
• CTA-608 CC1, in-band in every video track

The CTA-608 captions are signaled per video track with an accessibility descriptor (urn:scte:dash:cc:cea-608:2015, CC1=eng) in the MSF / CMSF catalog; catalogless moq-mi signals nothing and is in-band only. See the captions section.

🔒 Content Protection

Multiple content protection modes served as separate namespaces, each catalog carrying both CMAF and LOCMAF packaging

• mlm/cmsf/clear — unencrypted
• mlm/cmsf/drm-cbcs — Widevine, PlayReady, FairPlay
• mlm/cmsf/eccp-cbcs — ClearKey (ECCP)

Each namespace has its own CMSF draft-01 catalog, with every rendition in both CMAF and LOCMAF packaging. LOCMAF carries the same per-sample encryption metadata as CMAF, so the same EME / CDM path works for either.

Signaling follows the released CMSF draft-01 specification, also supported by Shaka Player.

📦 Packaging Modes

Side-by-side packaging modes so clients can compare wire-cost and pipelines

• mlm/cmsf/* — CMAF and LOCMAF (v0.3) in one catalog, sharing init data via initRef
• mlm/msf/clear — LOC, raw codec frames per object; video and audio only
• mlm/moq-mi/clear — moq-mi, catalogless interop; video and audio only

LOCMAF gives near-LOC overhead while keeping CMAF's metadata path, so DRM survives the round-trip. LOC and moq-mi carry raw codec frames played via WebCodecs.

⚡ Technical Specs

Modern streaming protocols and formats

• MoQ Transport draft-18
• ALPN-based version selection (moqt-18)
• AVC, HEVC and AV1 video; AAC, Opus and AC-3 audio
• CMAF chunks, LOCMAF delta moofs, LOC raw frames
• In-band CTA-608 CC1 captions (SEI for AVC/HEVC, T.35 OBU for AV1)
• Subtitles at the video object cadence, CMAF and LOCMAF, with paint-model stpc / wvtc
• Publisher priorities over raw QUIC: catalog, then audio and subtitles, then video
• MSE / EME and WebCodecs render pipelines in MOQ Player
• 10-second loops, wall-clock synchronized
• MSF / CMSF catalog per namespace
• moq-mi catalogless namespace
• Multiple namespaces announced from one publisher