Actually I just checked the same video Mirko linked above and when you right-click on the video and click “stats for nerds” it shows what streams are currently playing. For me it shows itag 251 for the audio stream, this means it’s 160 kbps Opus. That should be noticeably better than 128 kbps AAC. So it seems like nowadays it’s either 128 kbps AAC or 160 kbps Opus, depending on the video.

There isn’t any practical reason for “saving” 64 kilobits, given the size of the video stream, container overhead, and web bloat. I think YouTube wants to gently nudge users over to new formats for market control reasons. The MP4/720p format is/was all-round the best choice: compatibility, low data rate, low decode requirements, excellent quality downscaled by 2. I’d rather see all the lower qualities go. Today it is better to agree on one common format, instead of trying to save a kilobits at the cost of inconvenience. I very much dislike Google’s VP codecs, which have very short life span.

I believe that loudness normalizing is a good thing. It has been a feature of music players for more than a decade. No engineer is in control over the volume knob (I for instance keep it a notch down to allow for some headroom). However, since they are actually re-encoding the audio, it will result in quality loss. I’ve read that YouTube does keep the source upload archived for some time, but does not do so indefinitely, and eventually performs encoding from a reduced quality intermediate. This also explains the terrible quality of user uploads that are several years old, and have been transcoded for one reason or another.

I also welcome the switch to 48 kHz, which is the standard rate for video and PC equipment (since dvd / AC’97). I do wonder if there might be a double conversion going on, with YouTube internally still working at 44.1 / 16-bit, and feeding that to the Opus codec, and which is the best rate to upload at.

Hi Andrea! Yes, I believe that if you upload audio that is softer than the limiting threshold of YouTube’s audio protocol it will most likely bump up your audio to match everything else it has “normalized” across the site. I haven’t really been able to test this though so it’s just my best guess at this point.

Unfortunately, once YouTube has adjusted the level of your audio there’s not too much anyone can do about it. It will continue streaming to viewers at that volume until YouTube decides to change up the protocol again.

Every new upload still seems to retain a few days of “expected” loudness but this has actually seemed to have sped up since I last conducted this test. I haven’t checked exactly how long it takes nowadays but at the time I performed the test above, it was around a whole week before I noticed my upload audio change. Recently, I have noticed our uploads are affected in only a couple of days.

Wow, thanks for the super in-depth comment Nintendo Maniac 64! Based on your findings it would seem that there is really much more going on under the hood over at YouTube… and perhaps it would be safe to say that the use of several different formats is really more of an effort to keep viewer experience consistent over several platforms, as opposed to an effort to elevate or degrade quality.

With several billion views a day being the norm at this point, there are probably just too many variables to consider to stream a consistent format or codec at any given moment. Server speed, connection speed, internet browser, PC or tablet specs, etc. …the list just goes on and on and none of these are really even related to audio!

It’s been a while since I posted this article. Maybe it’s time for an update! I wonder if the normalizing is still in effect.

Aww crap I made a minor mistake – for 2014 and 2015 there was and is no “VP9/WebM” available on YouTube – there is only “VP9/DASH” for both 480p and 720p.

First off, for the last few years, YouTube has had two different video formats – AVC and WebM, the former being the incumbant format that is almost compatible anywhere while WebM is the newer format that only works on newer browsers, devices, etc. (YouTube technically has more formats for this, but they are only for 360p and lower resolutions and are only retained for fallback compatibility reasons)

Now in addition to these two formats, as you may know, YouTube encodes various different qualities and resolutions, and they do this for both formats; this applies to audio as well. The main difference for both audio and video formats are between 480p and 720p so I will be using them as a baseline for the comparison (in particular, everything 720p and higher use the exact same audio stream while everything 480p and lower use the exact same audio stream).

In the middle of 2013, this is what YouTube formats were for 480p and 720p:
-480p AVC/FLV + 128kbps AAC 44.1KHz
-480p VP8/WebM + 128kbps Vorbis 44.1KHz
-720p AVC/MP4 + 192kbps AAC 44.1KHz
-720p VP8/WebM + 192kbps Vorbis 44.1KHz

In the middle of 2014 this is what it looked like on YouTube:
-480p AVC/DASH + 128kbps AAC 44.1KHz
-480p VP9/DASH + 128kps Vorbis 44.1KHz
-720p AVC/MP4 + 192kbps AAC 44.1KHz
-720p AVC/DASH + 256kbps AAC 44.1KHz
-720p VP9/WebM + 192kbps Vorbis 44.1KHz

The main changes are that YouTube switched over to DASH streams (while still retaining the old 720p non-DASH AVC/MP4 for fallback reasons), replaced VP8 with VP9, and introduced a higher-quality AAC stream (it must be noted however that YouTube’s AAC encoder is not that great, so 192kbps Vorbis normally sounded better than their 256kbps AAC).

And as of today, in the middle of 2015, this is what it looks like:
-480p AVC/DASH + 128kbps AAC 44.1KHz
-480p VP9/DASH + 160kbps Opus 48KHz
-720p AVC/MP4 + 192kbps AAC 44.1KHz
-720p AVC/DASH + 128kbps AAC 44.1KHz
-720p VP9/WebM + 160kbps Opus 48KHz

Now YouTube has dropped 256kbps AAC, possibly because they now use the WebM formats as the default unless your browser/device doesn’t support it. Vorbis has been replaced by Opus but the 720p bitrate for WebM audio was reduced to 160kbps; Opus however is THE best lossy audio codec at this current time on Earth, so even at 160kbps the audio quality is comparable to Vorbis 192kbps and is definitely miles ahead of YouTube’s 128kbps AAC. One thing is that Opus as a codec is actually incapable of 44.1KHz which is why YouTube uses 48KHz; it should be noted that 48KHz is actually the standard audio sampling rate used in the broadcast industry, so this isn’t at all unprecedented and it should give a nice boost in quality for most modern audio which are usually recorded in 48KHz or 96KHz initially anyway.

It must be mentioned however that the old 256kbps AAC format can actually be found on some older videos even today, but new uploads do not have these formats; 192kbps Vorbis can also sometimes still be found on older video, but it’s very rare (much rarer nowadays than 256kbps AAC). Lastly, it must be noted that sometimes (but rarely) the Opus formats don’t even get encoded for some WebM encodes, resulting in them being stuck with 128kbps Vorbis at best.

To sum it up, this is the quality rankings I’d give regarding audio quality based on my own testing and others.

Audio uploaded with a sampling rate that is a multiple of 11025Hz:
256kbps AAC =* 160kbps Opus > 128kbps Vorbis > 128kbps AAC

Audio uploaded with a sampling rate that is a multiple of 8000Hz:
160kbps Opus > 256kbps AAC > 128kbps Vorbis > 128kbps AAC

(I left out 192kbps Vorbis because its extreme rarity makes it kind of pointless but also makes it very difficult to test; from my experience I’d say it’s at least equal to YouTube’s 256kbps AAC)

*they don’t actually sound the same but rather they both have their own pros and cons regarding the resulting audio quality

As a reference, 128kps Vorbis has commonly been compared to the likes of 192kbps MP3, but I wouldn’t be surprised if with modern versions of the LAME encoder that 192kbps MP3 would be more on-level with 160kbps Vorbis.

This is the YouTube downloader that I use which happens to support every single format they have (though as of this comment it seems to lack support for the lowest-quality Opus format, which is used for 144p):
http://cys-audiovideodownloader.com/

No worries Mirko! Thanks for explaining, I didn’t know you could access the debug info like that… Yes, 192Kbps is not that much of an improvement but at least there are options. Hopefully, this bitrate issue is just a temporary situation that YouTube is testing on its users. For the meantime though, I suggest everyone continues uploading their audio as best as possible so that if they do decide to bring back the higher bitrates, at least our uploads will also be affected.

In case you’re interested, I checked a little more and unfortunately I was wrong and youtube really always plays the 128 kbps stream. It appears that you can also check this by right clicking the video and copying debug info. There’s an fmt value which determines what video track is played and afmt which is the audio track. afmt always has a value of 140 (check the format codes in my previous comment).

I tried several youtube scripts and browser addons but none of them changes audio quality, only video, so no luck with that. I also checked Kodi (ex. XBMC) and a foobar youtube addon and it appears that they can play a 192 kbps audio track that is contained within the 720p format code 22 stream. I believe that’s the old higher quality audio that you would get when setting playback to HD. It’s not much, but I guess it’s somewhat better than 128.

Anyway, I’ve been wondering lately how come most of the music I was listening on youtube wasn’t going over some -4 dBFS, so thanks for the great article that explained that. 🙂

Hi Mirko, yup no problem and thanks for the link! I will check it out once I get a chance… as I said earlier though, I haven’t really had any opportunity to test the 128Kbps issue too thoroughly and I actually only recorded the streams from different resolutions just so I could hear the different recordings side by side and test whether Nick’s theory was correct. Unfortunately, they did sound the same to me as I switched from file to file across all resolutions. Furthermore, none of them really represented my original uploaded audio too well at all. To be honest though, once I realized the normalization was occurring I got pretty distracted and decided to test that out a bit more instead.

I am aware that YouTube has the data available for users to switch between, and I do know that I was able to upgrade audio streaming quality in the past. But I would have to agree with Nick on his point that as of this writing, the audio is not switching to higher bitrates regardless of the video resolution. Whether or not this is a resource hog on their end is really anyone’s guess!