Strange noises in NHL Stanley Cup Stream

In the Stanley Cup streams we’ve been watching, there are alternating faint tones in the background of all the game audio. They sound like “modem noises” and step discretely through a set of tones at a rate of about 2Hz.

I’ve been hearing them for several weeks now, and googling around it looks like other people are complaining too.

I did a little investigation and I can see them in the audio. See the horizontal lines at the bottom of this image? It’s divided into three of the distinct “notes” from the noise.

Filtering down to just this area (1-3kHz) they are extremely audible above the noise.
image

A frequency analysis shows that there are distinct groups of frequency peaks in each of these regions.

Note #1:

Looping the audio for the three “notes” as an example, you can see the grouping of peaks shift.

It’s extremely distracting, and it seems to be happening in a bunch of different streams (not just Hockey) on more than one platform.

I have three educated guesses:

  1. Audio artifacts from poorly configured noise reduction
  2. Audio artifacts from a messed up encoding pipeline
  3. Audio watermarking

I’m finding people complaining about this exact noise pattern in several different places across a number of streams.

The most complaints about it are on Sportsnet, across all sports. Interestingly it’s not there for any commercials: just the sport itself. It gets louder in proportion to the overall loudness of the stream but it’s the same basic pattern in all cases.

Okay, I might be going insane, but even at absolute full blast on every control I have, there is no audio on that youtube video. Doesn’t even show up on the EQ.

Could be hearing issues

Or YouTube cuts out the low frequency noises

1 Like

I can’t hear shit in the YouTube video, but audio watermarking does sound like a strong possibility. I know SSTV uses that audio range (though I doubt it’s SSTV). If you can isolate just the data sounds, try running it through something like FLDigi and see if it can make heads or tails of it?

If I were a betting man, I’d say five will get you ten it’s FSK-encoded data; it’s a cheap algorythm (both in terms of cost and computer power), and it’s robust against the kind of things that they’d want it to be robust against, like corruption and static.

I’ll take my EQ to the audiologist promptly.

Huh! Didn’t know that, and does seem a tad more likely.

Also, the bands definitely look like data streams. Here’s a ham radio signal ID wiki that might help: Amateur Radio - Signal Identification Wiki

8PSK seems to look the most like that spectrogram (the image here is a waterfall, so rotate 90 degrees clockwise mentally): 8PSK - Signal Identification Wiki

Hah it’s visual only. I made it real quick to show the bands and I didn’t have time to get audio simultaneously.

Agreed on that. I think I identified an initialization sequence that recurs periodically.

My signal isn’t the cleanest… I’m recording via OBS from a pirate stream of someone else’s stream via Sportsnet. I’m going to look for cleaner sources.

I can imagine an encoding pipeline being set up badly enough to cause this, and I can also imagine noise reduction being set extremely aggressively also causing it.

But the fact that I’ve since found a bunch of people who opened support tickets to Sportsnet and are getting the runaround… This could well be a poorly implemented watermark.

8PSK is pretty resilient (consider the noise floor on just about any of the ham frequencies), so you might be able to extract it with FLDigi.

I’ll try and report back the results!

Just make sure if you’re watching that the microphones on your cellphones are off, just in case it’s some kind of tracking nonsense.

In case YouTube keeps screwing with the audio, maybe you can just upload the raw video to your server?

A better isolation from analysis of a long clip.

1 Like

Identifying the specific tones used, I was able to make a series of notch filters (one for each) that is effective in removing the noise.

Top frequencies for each of three tones. Note that this was a granular analysis and a few of those are obviously the same in effect.

|1137.22|1297.38|2069.88|
|1135.88|1296.03|1096.85|
|1687.66|1671.51|1098.19|
|1686.32|1121.07|2071.22|
|1312.18|1298.72|2068.53|
|1313.53|2094.1|1273.15|
|1484.45|1119.73|1274.5|
|1138.57|2092.76|1095.5|
|1134.53|1672.86|1648.64|
|1483.1|2285.21|2664.73|
|1894.92|2683.58|1519.44|

1 Like
1 Like

I submitted this thread and YouTube video to the crew behind the Twenty Thousand Hertz podcast, who are experts in the field of sound. Let’s hope my query produces results.

1 Like

For those playing at home, you can use notch filters at these frequencies to eliminate the noise:

1119 Hz
1121 Hz
1136 Hz
1297.5 Hz
1313 Hz
1484 Hz
1686 Hz
1671.5 Hz
2092 Hz