stream verification · coming soon
Noisypeak Control
Watches your streams the way a player does and tells you what is wrong, with the evidence. Coming soon.
Noisypeak Control is the DigitalEye Stream Verifier, now in its final release before production. Give it a master playlist or an MPD and it does what a player does: it reloads the manifests, downloads the segments, probes and decodes them and extracts a frame per chunk. Then it tells you what a player never will: whether every tag, timing, bit rate, GOP, colour flag, loudness level and ad marker is what the specification and your platforms expect. When a check fails, it raises an alarm with the evidence: the manifest as it was, the segment record, the exact ffprobe command and its output.
Severity follows playback, not pedantry: a stream that plays fine on real players is never red. Yellow is a deviation players cope with, orange is really wrong, red is not playable or not reachable. One compact row per channel, a window per channel for the long watch, tiers per channel so a hundred channels cost what a hundred channels should. Launching soon as a hosted service with a 14-day trial; custom locations, SLA and on-premise on request.
It checks manifests, segments, decoded frames and audio levels, which goes well beyond pinging a URL.
Every raise keeps the manifest, the diff, the segment and the ffprobe output. Reproduce the finding yourself.
Severity by playback impact. Spec deviations that players tolerate stay yellow; nobody is paged for a missing tag.
Three tiers per channel; paused channels free the slot; boost a channel to full inspection for a launch and back.
One row per channel, one window for the long watch
The channel list shows one compact row per stream: name, what it plays (HLS v3 TS, HLS v7 fMP4, DASH fMP4 $Time$), tier, the tags the verifier detected (codecs, HDR, captions, DRM, ad markers), the health strip over the selected range, the ladder with its live-edge lag, and the last event. Filter by text or by tag logic, sort worst-first, act on many channels at once. Groups are saved filters.
Click a channel and it opens in its own window with Summary (what the stream is and who is behind it: packager, origin, CDN and ad stitcher, recognised from signatures), Preview (the latest frame of every rendition side by side), Renditions, Health, Stats and Events. To watch a few channels for hours, put several windows side by side.
100+ checks, each with its rule
Every check states what it verifies, what a failure means for viewers, where the rule comes from (RFC 8216, the DASH-IF timing model, Apple's HLS authoring specification or operational practice) and how it is evaluated: raised after N failing polls, cleared after M passing ones, with a start-up phase so a restarting channel isn't red for a minute. Groups: transport and HTTP, live manifest, HLS playlist, DASH MPD, segment download, MPEG-TS and fMP4/CMAF container, stream probe, video content, audio content, ad markers, rendition alignment, verifier health.
Among them: publish jitter and stale playlists, PDT gaps and drift, bit rate against BANDWIDTH and AVERAGE-BANDWIDTH, GOP and IDR regularity, colour signalling, interlaced sources, A/V skew, black and frozen pictures, silence and loudness (ATSC A/85 or EBU R 128 targets), SCTE-35 CUE-OUT/IN consistency and alignment to IDR frames, DASH timeline and UTC timing, AES-128 key fetch, and whether all renditions show the same moment.
- Thresholds per channel: tolerances, expected colour, loudness target
- Shun an alarm you know about: still recorded, no longer colours the channel
- Change-type alarms stay visible for an hour so a one-off change cannot pass unseen
Evidence you can reproduce
An alarm without evidence is an argument; an alarm with evidence is a ticket. Every raise and clear is an event; click it and you get the manifest as it was and the previous one with the difference, the segment record, and for segment-level alarms the exact ffprobe and ffmpeg commands that were run and their output. Send it to the encoder vendor or the CDN as it is.
Tiers, slots and boosts
Playlist reads every manifest on every reload and downloads nothing: availability and manifest hygiene for many channels at low cost. Adaptive-high also downloads the segments one at a time, like a player on a good link (mostly the top rendition, now and then a lower one), and probes and decodes them. Full does everything on every rendition and compares the first frame of each segment across the ladder. It is the only tier that proves the ladder is aligned. A slot licenses one active channel of its tier; pausing a channel frees the slot. Every period includes boost hours to lift any channel to a higher tier for a launch or a hunt.
- Playlist · Adaptive-high · Full, set per channel
- Slots per tier, prorated changes, paused channels free the slot
- Boost hours included; boost packs on top
- VOD playlists and static MPDs played once, twice or in a loop
Encoder fleet redundancy: NP-RFS
For Noisypeak Cell fleets, Control also carries NP-RFS: 1+1, n+1 and n+m redundancy for clusters of 2 to 100 devices, up to 20 clusters per instance. Each primary's configuration is replicated to central storage; when a primary fails its standby takes the channels, a publishing gate ensures primary and backup never publish the same output at once, and return to the primary is automatic or manual. Zabbix, Prometheus/Grafana and SNMP see the whole fleet.
Features
Reloads manifests, downloads and decodes segments and extracts frames, then checks every tag, timing, bit rate, GOP, colour flag, loudness and ad marker against the spec.
RFC 8216, DASH-IF, Apple HLS spec or practice: transport, manifest, container, content, ad markers, rendition alignment.
The manifest and its diff, the segment record, the exact ffprobe output. Reproduce it, or forward it to the vendor as is.
Severity by playback impact: yellow players cope, orange really wrong, red not playable or not reachable. Shun what you already know.
SCTE-35 in HLS tags, MPD events and emsg: CUE-OUT/IN pairs, durations, PDT alignment, IDR alignment. You see the problem before Prism X has to find out.
ATSC A/85 or EBU R 128 targets, true peak and silence, per rendition, continuously.
Full tier compares the first frame of each segment across renditions: the only way to know all renditions show the same moment.
Packager, origin, CDN and ad stitcher recognised from signatures, which helps when the stream isn't yours.
Playlist, adaptive-high, full; slots, boosts, pause. A hundred channels cost what a hundred channels should.
NP-RFS: 1+1, n+1, n+m failover for clusters of 2–100 encoders, config replicated, Zabbix / Prometheus / SNMP.
Specifications
| Streams | |
|---|---|
| Protocols | HLS (MPEG-TS, fMP4/CMAF, EXT-X-MAP) · MPEG-DASH (SegmentTemplate $Time$ / $Number$, SegmentTimeline) |
| Live and VOD | Live channels · VOD playlists and static MPDs played at real time or download speed, once or looped |
| Encryption | AES-128 HLS (key fetch and decryption checked) · DRM streams observed at manifest and container level |
| Ad markers | SCTE-35 via EXT-X-CUE-OUT/IN, EXT-X-DATERANGE, MPD Events, emsg |
| Checks | |
|---|---|
| Catalogue | 100+ checks in 13 groups, each with spec reference, raise/clear counts, evidence kind |
| Profiles | playlist · download · full · optimized built in; clone to customise |
| Levels | info · warning · major · critical, by playback impact |
| Per-channel settings | poll interval, tolerances, expected colour, loudness target, start-up phase, request headers |
| Evidence and history | |
|---|---|
| Evidence | Manifest + previous + diff · segment record · ffprobe/ffmpeg commands and output · segment bytes for major alarms |
| Retention | Samples 7 d · 1-minute rollups 90 d · events 90 d · long-term statistics kept |
| Service | |
|---|---|
| Tiers | Playlist · Adaptive-high · Full, per channel; slots per tier; boost hours |
| Access | Hosted at smon.noisypeak.com: e-mail or Google sign-in, 14-day trial (coming soon) |
| Custom | Custom locations, SLA, invoicing and on-premise on request |
| NP-RFS (Cell fleets) | |
|---|---|
| Redundancy | 1+1 · n+1 · n+m; clusters of 2–100 devices, up to 20 clusters |
| Behaviour | Config replication · publishing gate · automatic or manual return · Stay on Slave |
| Monitoring | Zabbix · Prometheus/Grafana · SNMP · JSON API; up to 50 encoders in one console |
Where it is used
Every delivered channel watched 24/7 at playlist tier; adaptive-high on the money-makers; ad markers verified before the platform complains.
Full tier for a day: ladder alignment proven, every rendition decoded, evidence attached to the acceptance report.
Bit rate over BANDWIDTH on two renditions since Tuesday, with the segments and the ffprobe output to send to the encoder vendor.
Platform feeds and partner channels: the Summary tells you who packages and delivers them, and when they break.
Talk to us about Noisypeak Control
We handle evaluation in the Noisypeak cloud, deployment on premises or in a public cloud, integration and support. One team covers the whole pipeline.
sales@noisypeak.com · +41 78 955 27 02 · Brochure · PDF · Stream Verifier (preview)