Orcasound · Salish Sea · 03 September 2026, 18:35 UTC
What the hydrophones actually heard
Every live Orcasound node over 24–26 August 2026, 3 calendar days and 9 node-days of audio, with a mechanical
resonance at 4,816 Hz identified and rejected. It accounted for
1,629 of 3,009 detections, and for the large majority of the
Southern Resident calls one node appeared to be producing.
88%of Bush Point's 625 Southern Resident detections were a metal object ringing, not a whale. 72 survive. The same filter removed nothing at Lab or North San Juan Channel.
25,079windows analysed
3,009detections with audio
1,629rejected as resonance
1,380detections remaining
What was run
3 hydrophones,
Bush Point, North San Juan Channel, Lab, over 24–26 August 2026, 3 calendar days and 9 node-days of audio:
25,079 thirty-second windows, 209 hours of audio. OrcaHello, a Southern Resident presence detector, flagged
3,009 of them as holding at least one positive segment, which is
12.0% of the audio. Every one was then classified for
ecotype and measured for the resonance.
Bush Point · Admiralty Inlet, off Whidbey Island · 48.03°N, 122.61°W · 69 h over 3 days
Lab · Haro Strait, west side of San Juan Island · 48.56°N, 123.17°W · 71 h over 3 days
North San Juan Channel · north San Juan Channel · 48.62°N, 123.09°W · 69 h over 3 days
Two counts exist and they are not the same. OrcaHello also returns
a whole-window verdict, which fires on 2,279 windows; 730
further windows hold a positive segment without it. This page counts segments
throughout, because a segment is what produces a clip to listen to.
Not included, because they are inactive in the feed registry rather than quietly missing:
Orcasound MaST Center · node offline upstream; acoustic signal stopped 2026-07-12, segments continued until 2026-07-17
Orcasound Point Robinson · hydrophone sediment-buried; the node streams but carries no usable signal
When each hydrophone heard something
Counts alone cannot tell an encounter from a fixed source. Per day they are
different objects. A pod passing a node is loud for an hour and then gone; a
machine is heard for a third of every day.
survives that
node's error ratenot
distinguishable from itnot recorded
Hours per day holding at least one ecotype claim, out of 24, on a
scale shared by all three nodes. Ticks below the baseline are days the node
recorded nothing, which is not the same as a day with no whales. Hover a bar for
its labels.
A named ecotype was claimed at Bush Point on 3 of its 3 recorded days, 6 to 10 hours on each — every day it was listening; Lab on 2 of its 3 recorded days, 1 to 2 hours on each. Southern Residents pass Admiralty Inlet roughly once a month, so a claim on every recorded day at a single node describes a fixed source rather than an animal. The embedding test below reaches the same verdict by a different route.
The 4,816 Hz resonance
Frozen Perch 2.0 embeddings put Bush Point's saturated hours in a single
tight non-biological cluster, 26.3 standard deviations tighter than true
background, that no other hydrophone contributed a detection to. That said the
audio was not whales. It did not say what it was.
It is a high-Q mechanical resonance: a fundamental at
4,816 Hz, 20.5 dB over true background, −3 dB width
62 Hz so Q = 78, carrying integer overtones at 2× and
3× and nothing at 0.5×. Across two days its centre holds to a
standard deviation of 87 Hz, 1.8%. It is struck rather than running:
15.5 dB in windows the detector fired on against 1.4 dB in windows it
did not. That is a metal object near the hydrophone, moved by water, ringing at
its fundamental mode.
The spectrum below is not a single line at 4,816 Hz, and the
claim does not need it to be. There is further raised structure between roughly
5 and 7 kHz, one part of it higher than either marked overtone. Struck
plates and bars ring in inharmonic modes as well as integer ones, so extra peaks
are what a metal object should look like rather than a problem for the reading.
What carries the identification is the narrowness of the 4,816 Hz peak
(Q = 78), the stability of its centre across days, its integer
overtones, and the absent subharmonic, not its being the only feature in
the band.
Mean spectrum of the windows the detector fired on, divided by the mean spectrum of true background, from 130 and 130 thirty-second windows. Flat at zero would mean the two sound alike. The fundamental and its integer overtones are marked; there is nothing at half the fundamental, which is what separates a resonant mode from a harmonic series with a missing root.
rejected as
resonancebelow the
threshold
Tone strength at 4,816 Hz, median with the 10th to 90th
percentile. The threshold is 18 dB, above the highest value
observed at either clean node. Hover a row for its counts.
A Southern Resident call is tonal with a harmonic stack in roughly
1–6 kHz, and so is this, which is why spectral flatness,
centroid, band ratios and pairwise spectrum similarity all failed to separate it.
The sound is not un-whale-like in its spectrum. It simply never changes, and none
of those measures were measuring change.
Before and after
keptrejected as resonance
The share of each hydrophone's own detections kept and rejected.
Each bar is that site's total, not a shared absolute scale: on a shared scale North San Juan Channel's 39 detections beside Bush Point's 2,925 would render as a sliver, and how little was taken from it is the finding. Counts
ride the bars.
Every detection each hydrophone produced against the audio it actually recorded, and what remains once the 4,816 Hz resonance is rejected. Greyed ecotype counts are not distinguishable from that node's own false-positive rate. These are counts of all detections, not of the clips played below.
Hydrophone
Recorded
Detections
Rejected
Remaining
Southern Resident
humpbackunder its error rate
Bush Point
69 h3 days, 8,306 windows
2,92535.22% of windows
1,62956%
1,296
72under this node's error rate
1under this node's error rate
Lab
71 h3 days, 8,498 windows
450.53% of windows
0
45
30unchanged
0
North San Juan Channel
69 h3 days, 8,275 windows
390.47% of windows
0
39
0
0
The threshold is calibrated on this sample:
18 dB sits about 1.3 dB above the highest tone measured
at the nodes where the object is absent. It is a filter for one object at one
hydrophone, not a general de-noiser, and another site's contaminant will sit
somewhere else in the spectrum.
Whether the numbers can be trusted
The ecotype classifier carries a measured false-positive floor on this audio:
across 40 windows OrcaHello scored negative it produced 3 spurious
Southern Resident calls and zero Bigg's. Every count here is
tested against 7.5% for Southern Resident and against a
Bigg's floor that 40 windows can only place below about
7.5%, using each hydrophone's own detections as the
denominator.
Removing the resonance does not make what
remains whales. It removes one identified contaminant. The
false-positive floor still applies to everything left, and a
no confident call verdict is weak evidence of absence rather than proof
of silence.
These labels are not distinguishable from the classifier's own error rate, so they are counted but not shown as detections. Each is tested against the false-positive floor for that label using the detections at that hydrophone as the denominator, not the pooled total across all three. The highest-scoring humpback is also the quietest audio in the sample, −66.4 dB against a background median of −61.1 dB, which is what a confident label on almost no signal looks like.
Bush Point · humpback: 1 of 1,296 detections, 0.08%, against a floor of 0.5% (6 expected from error alone; p = 1.00).
Bush Point · Southern Resident: 72 of 1,296 detections, 5.56%, against a floor of 7.5% (97 expected from error alone; p = 1.00).
Two limits on this test, both of which cut against reading it as proof. The floors are point estimates from small controls: Southern Resident is 3 of 40 windows, 95% CI 2.6%–19.9%, and Bigg's is 0 of 40, which is not a floor of zero but a floor no control this small can resolve below about 7.5%, so the Bigg's result is carried by the corroboration below and not by this test. And the test treats detections as independent draws, which a call bout is not: one encounter produces many correlated detections, which inflates any label that arrives in bouts. It is a screen, not a proof.
Does what survives hold up?
Rejecting the resonance is not the same as finding whales, so the detections that survive at Bush Point were tested separately. The tone measurement alone clears them: they sit at a median 2.7 dB over their local floor, lower than Orcasound Lab's confirmed whales at 7.8 dB.
Frozen Perch 2.0 embeddings disagree, and they are the stronger evidence because they share no training data, architecture or label space with the detector or the ecotype classifier. In embedding space the survivors sit nearest the rejected machine detections and furthest from the Lab's whales.
Mean cosine distance to the Bush Point machine cluster. Lower means more alike. Bush Point background is the control: same hydrophone, same water, same ambient noise, so if closeness to the machine were a property of the site rather than the source, it would sit as close.
Group
n
Distance to the machine
Bush Point survivors
24
0.4887
Bush Point background, same hydrophone
193
0.5758
Orcasound Lab whales
26
0.6445
The survivors are not exonerated. They sit 0.0871 closer to the machine than same-site background does, p = 0.0001 over 10,000 shuffles that exchange the two Bush Point groups (-3.9 SD of the null). The filter is necessary and not sufficient, and on this window Bush Point has no Southern Resident detection that survives scrutiny.
Only 24 survivors carry an embedding, so this is a small sample and the direction of the result matters more than its size. It also tests one alternative, the identified resonance, and cannot rule out a different contaminant.
Does the independent record agree?
A classifier agreeing with itself is not evidence. These are the days the
surviving detections concentrate on, set against what people on the water
recorded the same day. Nothing below was used to produce the detections.
Lab · Southern Resident
29 of 30 on 2026-08-26, of which 28 fall in one 33-minute bout, 14:22–14:55 local, and 1 sits elsewhere in the day. One further detection falls on another day.
Independently recorded that day: J and K pods, Southern Residents. “J- and K-Pods heading north towards the Strait of Georgia”, reporting “a large group of spread out whales inbound towards San Juan Island this morning”. Orcasound Lab sits on the west side of San Juan Island, on that route. · Orca Behavior Institute.
One non-corroboration belongs here too. A San Juan Safaris trip left Friday Harbor at 12:30 on 26 August, ran “San Juan Channel and around Henry Island into Haro Strait”, and logged a humpback with no killer whales. It does not refute the Orcasound Lab bout (that boat was working a humpback, and the pods were reported heading north), but a section listing only the confirmations would be the ranked-queue mistake again.
What survives, and what it sounds like
15 clips, drawn only from detections that
survived the filter and carry a named ecotype, best-scoring first within each
site. This is a listening sample and not a census, so read the table above
for how much of each site it represents. 15 of 15 clips are the full call bout rebuilt from the 30s source window.
Lab
30 surviving detections with a named ecotype
1.
30s
Southern Resident 1.00OrcaHello 0.984.8 kHz 11 dB
9 positive segment(s) · hour 21% positive
2.
30s
Southern Resident 1.00OrcaHello 0.974.8 kHz 13 dB
10 positive segment(s) · hour 21% positive
3.
30s
Southern Resident 1.00OrcaHello 0.914.8 kHz 14 dB
7 positive segment(s) · hour 21% positive
4.
30s
Southern Resident 1.00OrcaHello 0.934.8 kHz 7 dB
8 positive segment(s) · hour 21% positive
5.
30s
Southern Resident 1.00OrcaHello 0.924.8 kHz 5 dB
6 positive segment(s) · hour 21% positive
6.
30s
Southern Resident 1.00OrcaHello 0.974.8 kHz 2 dB
3 positive segment(s) · hour 21% positive
7.
30s
Southern Resident 1.00OrcaHello 0.964.8 kHz 15 dB
9 positive segment(s) · hour 21% positive
8.
30s
Southern Resident 1.00OrcaHello 0.974.8 kHz 11 dB
9 positive segment(s) · hour 21% positive
9.
30s
Southern Resident 1.00OrcaHello 0.974.8 kHz 15 dB
9 positive segment(s) · hour 21% positive
10.
30s
Southern Resident 1.00OrcaHello 0.954.8 kHz 6 dB
7 positive segment(s) · hour 21% positive
11.
30s
Southern Resident 1.00OrcaHello 0.984.8 kHz 7 dB
9 positive segment(s) · hour 21% positive
12.
30s
Southern Resident 1.00OrcaHello 0.974.8 kHz 12 dB
11 positive segment(s) · hour 21% positive
13.
30s
Southern Resident 0.99OrcaHello 0.974.8 kHz 3 dB
7 positive segment(s) · hour 21% positive
14.
30s
Southern Resident 0.99OrcaHello 0.994.8 kHz 10 dB
10 positive segment(s) · hour 21% positive
15.
30s
Southern Resident 0.99OrcaHello 0.974.8 kHz 17 dB
13 positive segment(s) · hour 21% positive
Limitations
OrcaHello gates everything downstream. 12.0% of this
audio reached a classifier at all, and a Bigg's call has to sound
Southern-Resident enough to trip a Southern-Resident detector.
The rejection threshold is one number, calibrated on 3,009 tone
measurements across 3 days at three nodes, with roughly
1.3 dB of margin. It is a filter for one object at one hydrophone.
The false-positive floors are measured on windows OrcaHello scored
negative. Those windows are quieter and less tonal than the windows a
detection comes from, so the floors probably understate the rate that
applies to the audio actually being labelled.
The same ecotype classifier recovered only 5 of 20 killer whale recordings
from the Watkins database, which is expected for a model specific to
Salish Sea repertoires and means absence is weakly evidenced.
These are the three live Orcasound nodes. MaST Center and Point Robinson
are dead upstream and are listed above rather than omitted.
These pages, the code that produced them and this prose are written with AI coding agents under human direction, and reviewed before publication. How these reports are produced.