Sound Problems: high pitch buzz

Course Q&A Deep Learning Specialization Improving Deep Neural Networks: Hyperparameter tun
1

Hi Deep Learning Specialization team!
I’ve been taking many classes from your DeepLearning.AI series, but this one in particular ( Improving Deep Neural Networks) has a constant high pitched buzz that’s been consistent throughout all videos (I’m already halfway on week 2). It may be a bit imperceptible to people with not so great hearing (not meaning to offend anyone, but as we grow older we lose our ability to hear higher pitched sounds). I’d recommend the team to pass a high pitch filter or ask an audio engineer to help with fixing this.

Thank you in advance, and keep up the great work!

2

Hello,

I just watched a few random videos (at random times) of week 2 of this DLS course 2. But I didn’t notice any high pitch buzz. Can you please give me links of these videos with time frame where you notice the high pitch?

Best,
Saif.

3

I believe you can run an equalizer on your computer to apply additional filtering.

4

Sure, happy to point to specific examples. But please note that it is actually happening across most videos (>80%).

Note: It may help you to notice it if you use headphones instead of using your laptop speakers.

Video examples

  1. Bias / Variance (second video of Week 1). High background pitch from the start of the video at t=0. Remains throughout all the video
  2. Learning Rate Decay (second to last video of Week 2). High pitch noise accentuated whenever Andrew talks

Example of a video that DOESN’T have the high pitch problem

  1. The Problem of Local Optima (last video of Week 2). You’ll see there’s no buzz here.

Hope this helps identifying the issue. Let me know if you need any more details.

5

Thank you for the details. I highly appreciate your time and efforts.

Using hands-free, I didn’t notice any high pitch. Just notice a small buzz barely for a few seconds. But maybe this problem is with headphones or with your headphones only. I don’t have headphones but if you have multiple, can you try another one and posted us?

Alternatively, as Tom mentioned, you can run an equalizer for filtering. Actually, I didn’t hear any learner complaining about this issue, and neither I can see this issue. Only you (I guess) faced this. So, frankly speaking, if we forward your suggestions to concerns, there will be no action as no mass is facing this issue.

But thanks a million for your time and suggestion. I highly appreciate your feedback. Please let us know if you have any additional feedback or concerns.

Best,
Saif.

6

I re-checked the Bias / Variance video with a good-quality headset.

The audio certainly isn’t studio quality, but not bad for a guy with a lapel mic in an acoustically live office.

I didn’t hear any high-pitched buzz.

7

I also didn’t hear any high pitch buzz. Its audio quality is not as great as new MLS but its good to me.

8

First, thanks for being receptive to the feedback!

OK, now, I honestly think it is very likely that some of you have already lost the ability to listen to these high pitched sounds (maybe ask your kids to listen and see if they can). The buzz is quite mild, but it’s there! (I can definitely tolerate it and make out all the content, so all good on that front; it’s just a bit annoying).

Just for you to consider: I’ve tried using different headphones: wired ones, bluetooth ones, high end ones (expensive), low end ones, and even tried using different computers (a mac and a windows), all with the same result. So, unless it’s only me having weird ears, I do believe the buzz is there.

Now, if I’m the only one noticing it, maybe it’s not worth the hassle trying to fix it! As I said, the content is perfectly understandable as is.

9

I’m also hearing this high-pitched noise. I downloaded the Bias and Variance video and did a quick spectral analysis, there is a tone at about 11.3kHz.
Playing around in an audio editor it can be removed with a notch filter at this frequency or even some more advanced AI tools can improve the audio (Adobe’s Podcast AI is a great tool for this: Enhance Speech from Adobe | Free AI filter for cleaning up spoken audio)

image
image900×749 38.7 KB

10

Interesting plot. I’ll forward that to the course staff.

11

Hi @jojasan

In my personal experience, adjusting the video quality can sometimes impact the audio, so experimenting with different qualities might help mitigate the issue.

12

I recommend you asking a kid to listen to the recording… The case might be that adults lose the ability to listen to these frequencies…

13

The audio analysis plot is pretty definitive.
As mentioned, it’s been reported to the course staff.

14

I hear it too. Bad mic and or bad codec. I am pretty sure I have hyperacuity. Love the content, but the audio quality is killing me.

1 Like
15

In point of fact, this looks and sounds like aliasing noise. Something in the backend, perhaps the codec, is improperly resampling the audio without a sufficiently powerful anti-aliasing lowpass filter.

1 Like
16

It sounds like (pun intended) you have a very sophisticated understanding of audio technology. Perhaps you can add that filter in your output stream until the course staff gets around to addressing the issue.

I’m of a certain age (got the grey hair honestly) and haven’t had my hearing checked in a few years, but I’m pretty sure I don’t hear much of anything above maybe 8khz these days so it sounded ok to me. :nerd_face:

17

Hah, well I do hold a PhD in acoustics and have spent the last decade as a DSP engineer (ML is strikingly close and I am loving it!)

I don’t have the best hearing as a lifelong musician approaching 40 (I know I’m still a ‘kid’), but when I see a plot I know what’s going on.

On PCs the Peace equalizer can be used. It’s not so cut and dry because cutting high frequencies reduces sibilance and therefore speech intelligibility. So you can cut out some of the noise…but you are also cutting out some of the signal. Peace Equalizer, interface Equalizer APO / Peace Wiki / Features FAQ

More sophisticated post processing tools can be used such as iZotope Rx but the have an expensive analysis step that cannot be used ‘live’ without compromising lip sync.

2 Likes