Dek: A broadcast audio-silence alarm should distinguish a missing program feed from legitimate quiet before it pages master control or triggers a source switch.
By Kyle Suess | Engineering field note | November 20, 2026
An audio PID can remain present while the audience hears nothing. Equally, a healthy program can contain a silent scene, a pause in commentary, or a deliberately muted microphone. Treating all three as the same incident creates alarm fatigue and can switch a healthy service to an equally silent backup.
Why it matters
Audio presence, decoded level, program meaning and source health are separate questions. An MPEG-TS continuity check can pass while decoded samples are silent. A loudness measurement can report a low value without proving a technical failure. A speech detector may find no dialogue during a perfectly healthy music or ambient-sound segment.
AWS MediaLive's automatic input failover documentation distinguishes input loss, black video and audio silence as separate failover conditions. Its AudioSilenceFailoverSettings reference defines silence for that feature as audio loss or audio quieter than -50 dBFS, sustained for a configured duration. That threshold is a specific product behavior, not a universal broadcast silence standard.
Six checks before escalating a silence alarm
1. Verify the intended audio service. Check program number, audio PID, codec, channel layout and language selection. A monitor following an unused commentary or alternate-language PID may correctly report silence while the main program is healthy.
2. Separate transport from decoded audio. Check packet continuity and decoder errors independently of sample level. An intact compressed audio stream can carry digital silence; missing transport packets may instead cause decoder concealment or muting.
3. Inspect channels independently. Measure left, right and multichannel components before reducing them to a single number. A silent center channel can matter for dialogue even when other channels contain ambience. Also check whether a downmix masks a missing source.
4. Apply time and context. A short pause in speech is not equivalent to minutes of unexpected dead air. Set thresholds using the service type, scheduled content and operational risk, not a single value copied across every feed.
5. Correlate the upstream and downstream points. Compare the contribution input, decoder output, playout output and final distribution rendition where available. If audio exists before an encoder but not after it, the likely fault domain is narrower than a generic channel-down alarm.
6. Validate the backup before switching. A second source may share the same upstream silent mix. AWS documents that automatic input failover operates on an input pair, while manual forcing of failover can be appropriate when operators detect degradation before the automatic condition fires. Neither proves the replacement carries useful audio.
What operators should change
Build an incident record with the selected PID, codec, channel layout, measured levels, detection duration, transport errors, current source, backup source and correlated downstream symptoms. Keep the raw observations even when several alerts collapse into one operator incident.
Use at least two alarm states: a lower-severity audio below threshold observation and a higher-severity unexpected service silence condition supported by program context or additional evidence. Document which states may initiate automatic failover and which require operator review.
Test against a short dramatic pause, an extended silent segment, a missing audio PID, valid encoded digital silence, a failed decoder and a backup carrying the same fault. Measure time to detection, false alerts and the result of source switching. These are suggested acceptance tests, not claims about a particular broadcaster's deployment.
AWS's failover and failback behavior also shows why recovery deserves testing: a source becoming healthy again does not necessarily mean the channel will immediately return to it. Alarm-clear rules should not silently assume that failback occurred.
The engineering takeaway
The question is not merely whether the meter reads near zero. It is whether the correct audience-facing service has lost expected sound, where that loss began, and whether switching inputs will improve it. Build alarms around that decision and preserve enough evidence for an operator to act.
Image credit: Abigail Keenan, [Mixing console](https://commons.wikimedia.org/wiki/File:Mixing_console_(Unsplash).jpg), 2016, CC0 1.0. Archival equipment photograph, not an image of a current incident. Intended publication derivative: cropped/resized and JPEG re-encoded.



