Piano World Home Page Forums Our Most Popular Forums Digital Pianos - Electronic Pianos - Synths & Keyboards YouTube Audio Fidelity test -- help!

I have often seen people write:

. . . "Based on what I've heard on YouTube . . . ".

But I don't trust YouTube's audio fidelity.

So I've made an audio file with some white noise, and some pink noise.


I'd like to upload it to YouTube,

. . . and then play it back through YouTube,

. . . . and record the audio playback as a WAV file, so I could analyze the playback with Audacity.

A comparison of the input spectra, to the output spectra, could be informative.

But I know _nothing_ about videos -- and YouTube won't take audio files.

I don't care about the "video" part -- plain white would be fine.

The URL for the input audio WAV (it's a Dropbox file -- this may be troublesome):

https://www.dropbox.com/s/pl4ki78wq9tykj2/YouTube%20White%20Noise%20v2%20HPF.wav?dl=0


If someone could do this for me, or give me a pointer to a "How to Create Video Files for Dummies" page, I'd be grateful.

The noise has been put through a high-pass filter -- flat down to 30 Hz, sharply fading to -60 dB at 20 Hz.

Thanks --

There really is no need to do this.
Encode your file as 192 KBit aac and you should have a pretty good picture of what you will get from a YT video.

It is not perfect but it is "good enough" for most cases, IIRC it is the same bitrate of multichannel Audio in DvD movies.

Ok so I fired up Sony Vegas and created a file.

First: direct link to the file I uploaded.
https://mega.nz/#!V4Jh1I7S!46Qyor84VIEDwb0ZAO4_JuTq0XislTM-xalhrXNtyh8
I can't render the video with uncompressed wav, so I compressed to 512KBit AAC (highest setting possible).

Second: Youtube link to the file (just DL the vid with any program of your choice, JDownloader2 will do so easily)


Keep us in the loop about your findings, will you? ^^

PS: Did I ever mention that the time limit on editing a post in this forum is a retarded concept? Get rid of it already.

This page provides some handy info on any given YouTube video, including links to direct componemts of the stream (e.g. just the audio) directly:

https://www.h3xed.com/blogmedia/youtube-info.php

Here's a direct link to the .m4a file in the video above.

It is indeed AAC, but with a bitrate of 128 KBit.

I guess it should be possible to load this back into Audacity for analysis (with directly with codecs, or converting back to WAV first.

Cheers,
James
x

I don’t know what’s your end goal with this experiment, but keep in mind that whatever you end up doing, YMMV in comparison to this experiment, because YouTube’s encoding is not the only factor:
It starts with the encoding of the source video. Granyala now used 512kb AAC, but a different source video with a different compression will give different results.
And it ends of course with the device that is used for listening to the YouTube video. An anecdote about this: I recently noticed that my Amazon Fire 10 tablet (which I use mainly to display sheet music) auto-normalizes the audio of YouTube videos, at least when using headphones. That has the effect that quieter parts of the audio become louder, greatly reducing the dynamic range and essentially ruining any piano piece that has pp/p parts as well as f/ff.

All good points JoBert.
I wonder whether normalization is done in the app. Do you get normalization if you watch a YT vid in a browser window too?

Be aware that that in the best case scenario, a YT stream is compressed twice. The first step of compression happens when encoding the source material as audio + video streams to be uploaded to YT. YT sets a limit to the compression rate that can be used in the uploaded files.

The second compression step happens at YT. The stream is decoded and it will be re-encoded with a YT compliant format. The best audio quality you will get is 192 kbps AAC. Usually it is 128 Kbps AAC. This encoding step is also not lossless.

This means that with YT the end result is always worse than converting the source directly (i.e. in a single encoding step) to 192 or 128K AAC.

Finally, only HQ video will get 192K audio. SD video will get a maximum of 128K. So, even if the goal is to upload audio to YT (like with a music album), the video stream should be HQ (720p or higher). I am not sure what YT is doing regarding audio with 4K video.

More info here
https://support.google.com/youtube/answer/1722171?hl=en&topic=2888648&ctx=topic

Doesn't YouTube support uploading videos with FLAC-encoded audio?

If so, the (lossy) compression would only occur once.

Kind regards,
James
x

According to YT's documentation, no. FLAC and other lossless audio formats are not referenced as supported formats.

Note that a FLAC audio stream can be added to an MKV container, but not (directly) to an MP4 container. MP4 containers are the ones supported by YT. It is possible to add lossless audio streams to an MP4 container, such as ALS (which is quite similar to FLAC) and Apple's ALAC. The question is whether youtube would decode those streams. They recommend uploading the audio at compressed at 384 kbps.

Yes, that's where I noticed it, using Amazon's own Silk browser. I don't have another browser on the tablet, so I couldn't test if this happens in all browsers, but it does happen with the YouTube app from the Amazon app store too. (I made lots of comparisons using my iPhone, where this does not seem to happen, to be sure that this effect is there. It definitely is).

I think the word "normalization" I chose isn't actually correct, as normalization only affects the max level (I think). I guess it's more an automatically adjusted decrease of the dynamic range. I guess that the reason for this is, to be "friendly" to the listener. If you surf through various videos that all have different volume levels, and you get to one with a low volume level, you may be tempted to up the master volume of the tablet to hear it better. If the next video is then one with a high volume level, you may blast your ears off before you manage to correct the master volume down. So for that purpose, it is probably not a bad idea. But for listening to classic music, it is really bad.
Well, I don't really use my table for this (as I said, it is mainly for sheet music). And if I do, then usually just for a quick listen as a reference during my practice ("how is this passage supposed to go?"), so it doesn't really affect me much, but it still was a surprise when I noticed it. (I noticed it while listening to one of my own YouTube videos, where I wondered "Did I really mess up the pp section that much? It doesn't even sound like p, more like mf!". I was relieved when via the iPhone I could hear a proper pp. )

Granular --

Thank you!

I'll install your download program, and report.

I will also try it with virtual audio cables, to see if the results are different.

What started me down this rabbit hole was doing some Mic tests on a large djembe. What I heard live was _way_ better than anything on Youtube.

This area is way more complicated than I realized. Thanks for all the other comments; they may lead to improvements in technique.

You are confusing two different practices.
Dynamic compression and volume normalization.

Dynamic compression = making quiet stuff louder, so the differences within the piece are smaller. Good for noisy listening environment or watching a movie at night (speech will be almost as loud as the effects, so you don't have to hastily adjust the volume everytime sth. goes boom )

Volume Normalization (also known as replay gain) = approximating the volume of different pieces. If done incorrectly this will lead to nasty clipping but if done correctly and provided there is enough headroom in the piece (a.k.a. FFF is not already 0dB), original dynamics remain intact.

IIRC Youtube only does the latter.

As this thread's topic is "audio fidelity", I will more explicitly write what others are stating: multiple lossy compression passes destroy audio fidelity in an unrecoverable fashion.

Different internet audio services may provide better "audio fidelity" than YouTube does, albeit with smaller audience reach and different business models.

Regarding normalization. Youtube does normalize the audio on the video streams. If you go to youtube, right click on top of a video that is playing and select "Stats for Nerds", you can see the normalization level being applied. Check the example below.



The first percentage (100%) is the volume setting in the youtube player. The second (47%) is the normalization value. The content loudness is the difference between the audio stream level and the reference level that youtube is using. In this example, the the audio is 6.6dB louder than the reference level. This means that youtube is reducing the volume level 47% (i.e. -6.6dB) to normalize it against the reference level. AFAIK, the maximum normalization value applied by youtube is 100%. This means that normalization only applies to audio above the reference level. Videos with loudness below the reference level will not be normalized and played louder.

(I really hate the edit time limit) Another side effect of normalization is that much of the high-visibility content on youtube (and other platforms as well) has the dynamic range tweaked so that it always sounds as loud as possible without running into audio level reductions. This applies to most of the "commercial" music and of course most adverts on those platforms. So, putting "audio fidelity" and "audio/video streaming platforms" on the same sentence makes no sense...

Well, I'm pretty sure that a music streaming service like Tidal does not compress further by itself.
Jack [censored] they can do if the recording is compressed to a brickwall during mastering though which, as you stated, is the case for most popular modern music. :X

You only get FLAC or ALAC lossless with Tidal if you pay 20 USD/month for the hifi subscription. With the other paid subscription options you get 320 or 256 Kbps encoding.

YouTube _does_ support the upload of uncompressed audio for some users.


These capabilities seem to be reserved for the owners of copyrighted material, who have established a relationship with YouTube. "Music videos" are a special case.

See here (and in several linked pages):

https://support.google.com/youtube/answer/4603579?hl=en


I _knew_ this was going to be complicated . . . .

Data rate compression is not the same as dynamic compression.
Naturally most streaming services compress data, we were talking about the compression of the musical dynamics, which is often done in the mastering process to achieve more loudness (the average Joe Tonedeaf apparently thinks that is a good thing).
You will have the same squashed dynamics whether you listen to a crappy 128kBit .mp3 or a 192KHz 32Bit uncompressed PCM file.