feat(stats): report audio echo-cancellation metrics in publisher stats - #3732
Open
raviteja83 wants to merge 3 commits into
Open
feat(stats): report audio echo-cancellation metrics in publisher stats#3732raviteja83 wants to merge 3 commits into
raviteja83 wants to merge 3 commits into
Conversation
The audio media-source stat carries echoReturnLoss and echoReturnLossEnhancement, the only signals that say whether the browser's echo canceller is running and how much echo it removes. media-source was previously resolved for video tracks only, so neither field reached publisher stats. Reports min/max per sample window rather than an average: ERLE legitimately drops to 0 during double-talk and silence, so a window mean makes an idle canceller and a working one look alike. erle_distinct_count is 1 when the value never moves across the window, which separates "no software AEC running" from "AEC running". Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
…ure level Audio tracks resolve a media-source for ERL/ERLE, which set sourceStatsAvailable and let the video-only frame-counter branch run on audio samples. Those counters coerce a missing value to 0, so every audio sample shipped source_total_frames: 0. Gate that branch on track kind instead. Capture energy was gated behind ERL/ERLE being present. ERL/ERLE only exist while a software canceller runs, so a track published with echoCancellation disabled lost its energy figures too — the cohort where "was the mic live at all" is the question. Report energy independently, and omit it when the browser never reported the counter so an absent metric is not indistinguishable from measured silence. sourceAudioLevel was collected but never aggregated into a sample. Report it as a range; no average, since RMS is already recoverable from the energy counters and integrated rather than sampled at poll time.
…tric docs differenceIfReported guarded the whole window while the subtraction reads only two values: this window's last stat and the previous window's last stat. A counter that stopped being reported mid-window emitted a fabricated 0, and a media-source replaced by a device switch or plugin toggle emitted a large negative — which divides against a negative duration to a plausible-looking RMS, reporting a loud mic during silence. It now guards the two values it actually reads and treats a backwards counter as unknown. The publisher and subscriber sample windows ignored the value plumbed in from init config and compared against the module constant instead, so maxSampleWindowSize could not take effect. Both now use the configured size, and the payload reports it rather than the default. buildAudioSourceStats no longer sets sourceStatsAvailable — that flag gates the video frame counters, and setting it for audio required a compensating gate that also narrowed video stats arriving without a kind. Audio media-source matching now accepts mediaType alongside kind, matching the video helper, so an engine reporting only mediaType does not silently lose every audio metric. Adds erle_observed_count and audio_level_observed_count: a single observation scores erle_distinct_count 1, previously indistinguishable from a canceller pinned across the whole window. Drops the toFixed(3) rounding, a 0.001 dB tolerance defending against a hazard three orders of magnitude below the smallest meaningful ERLE change. Corrects PublishAnalyticPayload.video, which pointed at LocalAudioTrackAnalytics and so asserted video tracks carry ERLE; the erle_distinct_count doc, which read 1 as "no canceller" when that case has no ERLE fields at all; and the ERL/ERLE doc, which inferred where cancellation happened from an absence the spec does not license.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds audio capture metrics to publisher stats from the audio
media-sourcestat, which was previously resolved for video tracks only:echoReturnLossandechoReturnLossEnhancement(the only signals indicating whether a canceller is applied to the capture path and how much echo it removes), total audio energy and samples duration, and capture level — each reported as a range with an observation count.ERL/ERLE are reported as min/max per sample window rather than an average, since ERLE legitimately drops to 0 during double-talk and silence and a window mean makes an idle canceller indistinguishable from a working one.
erle_distinct_countis 1 when the canceller reported a pinned value, read alongsideerle_observed_countbecause a single observation also scores 1. Energy is reported independently of ERL/ERLE so that tracks published withechoCancellationdisabled still carry capture evidence, and every counter delta is omitted rather than emitted as 0 or as a negative when the browser stops reporting it or the media-source is replaced. No level average is reported, since RMS is recoverable from the energy counters.Also fixes the publisher and subscriber sample windows, which compared against the module constant and so ignored
maxSampleWindowSizefrom init config.