There's a lot of confusion about what sample rate to use for YouTube.

YouTube's own guidelines for music videos
Bit depth: 24 bit is recommended, 16 bit accepted
Sample rate: 44.1 kHz is recommended, higher rates accepted
Channels: Stereo
Format: PCM (WAV or AIFF)

If you insist on delivering compressed audio the guidelines state:

Codec: AAC-LC
Sample rate: 44.1Khz
Bit rate: 320kbps or higher, 256 kbps accepted
Channels: Stereo

(source: Encoding specifications for music videos - YouTube Help)

YouTube's own guidelines for other or general material
Codec: AAC-LC
Channels: Stereo or Stereo + 5.1
Sample rate: 96khz or 48khz

Other formats are clearly accepted even though it's not stated, but the above is the official recommendation. Apparently YouTube doesn't want uncompressed audio here, which means transcoding is bound to happen. D'oh!

Other tests I've done don't show any penalty to the video quality if you upload full quality PCM audio instead of the recommended AAC, so there you go. I did this test to see if the total size of the video (i.e. the audio contribution in terms of file size) would compromise the video quality.

(source: Recommended upload encoding settings - YouTube Help)

Test
Obviously there's some potentially conflicting information so I decided to test what actually sounds or measures best, and what happens at different source sample rates.

Test files
· 24 bit PCM WAV
· 20 Hz to 20 kHz slow sine sweep generated at the actual target sample rate
· Identical sample peak, true peak and integrated loudness, headroom of 6 dB
· Accompanying video is 1080p HQ with no lossy compression of audio.

Why a simple sine sweep? It covers the full frequency spectrum and it's easy to detect distortion or aberations. Actual music might mask problems in the conversion file (which could be a good thing in real life, though).

Maybe I'll do a square wave test as well to see what happens with inter-sample peaks and how this could also affect the loudness reading slightly in a negative fashion.

Captured audio
· Converted audio from YouTube was captured digitally
· Converted audio was sharply low pass filtered @ 15 kHz by YouTube
· Converted audio was lowered in level by YouTube to match YouTube’s loudness normalization target

System playback @ 44.1 kHz

Results
Uploaded source: 44.1 kHz
Aliasing artifacts: None
Other noise: Very low noise around the AAC noisefloor @ 20-500 Hz
True Peak level: -14.93 dBTP
Sample peak level: -15.01 dBFS
Max. momentary loudness: -13.1 LUFS (I) (loudest in test)
Total RMS level: -16.52 dB (loudest in test)

Uploaded source: 48 kHz
Aliasing artifacts: Yes
Other noise: No
True Peak level: -15.63 dBTP (lowest true peak in test)
Sample peak level: -15.69 dBFS (lowest sample peak in test)
Max. momentary loudness: -13.2 LUFS (I)
Total RMS level: -16.53 dB

Uploaded source: 96 kHz
Aliasing artifacts: None
Other noise: No
True Peak level: -14.87 dBTP (highest true peak in test)
Sample peak level: -14.93 dBFS (highest sample peak in test)
Max. momentary loudness: -13.2 LUFS (I)
Total RMS level: -16.53 dB

Conclusion
It's hard to say what will sound worse on YouTube, but the 48 kHz conversion is the only one that has aliasing artifacts, which is a big no-no in my book. The 48 kHz version also features the lowest peak values after conversion for some reason.

Distortion is almost similar between the 44.1 and 96 kHz files, but while the 48 kHz file seems to have fewer problems in the low mids it has distortion in more areas in the high end.

On the other hand the 44.1 kHz conversion was the only one to feature some measurable noise around the AAC noisefloor, approximately at -100 dBFS, which I would expect to either be in all or none of the conversions.

For now I think it's safe to conclude at if you've mastered a track with a target rate of 44.1 kHz for digital aggregation or CD there's no point in doing a specific 48 kHz version for YouTube, it could even be detrimental to the sound.

However, a 96 kHz version shouldn't be bad if you're already working at that sample rate, but upsampling later won't help.

Disclaimer and other practical considerations
Here's the catch:

Many or most video editors by default work in 48 kHz sessions and will import a 44.1 kHz master into that session, automatically converting it to 48 kHz using what ever crap SRC is built into the video software. It's lazy, but that's the way it is.

So ironically you might be better off doing your own high quality 48 kHz SRC using e.g. Izotope RX7 for the video guy/gal even though it's likely worse for YouTube in the end. A better solution would be to educate the video editor and label, if you have the guts, but then there's VEVO...

VEVO is apparently different. Specs here are 44.1 kHz for AAC, but 24 or 16 bit 48 kHz for PCM, with Little Endian byte order during video export. So there's no way around a 48 kHz version for VEVO if the specs are to be believed, i.e. the video might be rejected on upload if it's PCM and not 48 kHz.

Pics

44.1 kHz sine sweep after conversion


48 kHz sine sweep after conversion, notice the aliasing lines in the background not present in the other conversions


96 kHz sine sweep after conversion

Download or open in new tabs to blow up details.

Thanks for the insight, good stuff!

Which is exactly what I do. If SRC results in overs, I SRC an unlimited master and re-apply the limiter settings @ 48khz. Usually I also lower the ceiling a little further.

Many people editing videos also seem to be completely unaware of what sample rate they are using. I have battled this many times..

It's all going to play through an iPhone anyway. At 48 kHz. ;-)

Thanks for this Holger!

Thanks for the detailed insight Holger!
My experience with this subject is that 44.1khz 24bit is the best one for streaming services, after extensive listening.

Many streaming services don't accept 24 bit files yet. So 16 bit is still necessary in many places.

The difference between a dithered 16 bit master and a 24 bit master is negligible, but a different sample rate can wreak havoc with the material in comparison.

This is great Lagerfeldt!
I've been meaning to research this for a long time - you saved me the trouble.
thanks!

C

Lagerfeldt, What video codec did you use?

It's most common for video editors to use h.264 for YouTube, which doesn't even have the option to contain lossless wav audio, only AAC or mp3 (at least it doesn't have this option in Premiere...)
Thank you!

The H.264 standard does include the option for lossless audio.

In fact I did several H.264 encoded (test) videos with lossless audio, but Apple ProRes 422 is my preferred codec after extensive comparisons.

As you correctly point out, for some reason Adobe Premiere Pro doesn't have an option for lossless audio with the standard H.264 export, only with H.264 for Blu-Ray via Adobe Media Exporter. I guess audio isn't their strong side.

I use Final Cut Pro X and Apple Compressor, which allows for very detailed export settings and encoding tweaks, including H.264 encoding with lossless linear PCM audio.

Thanks Lagerfeldt!

Does this mean that YouTube re-samples everything to 44.1 kHz ?
If yes, it only makes sense that 44.1 kHz file will sound better eventually.

I also assume that the audio is being converted by YouTube to a compressed AAC or mp3..?
In this case, I wonder what's better-
A. To upload the best sounding WAV file and let YouTube compress it, or
B. To try and find out what are the exact specification of the compression YouTube uses, do it myself, and hopefully the audio will be left untouched by YouTube.

Option A is the only real choice.

YouTube will always encode your audio.

This potentially leads to transcoding (=double encoding) if your audio is already lossy encoded, regardless of how closely you match YouTube's own settings in your uploaded video.

AFAIK YouTube still uses a variety of audio codecs for different platforms, including AAC, Opus, Vorbis, and even MP3. Anything you give YouTube will be chewed up and spat out in these formats depending on what platform is used to play the video.

Thanks a lot Lagerfeld ! I've been struggling with this for a long time.... Like mastervargas said you saved me the trouble.

I'm still wondering about 24 vs 16 bits for Youtube?

With -100 dBFS noisefloor it seems that we are closer to 16 than 24 bits dynamic. To my mind it's still useful to upload 24 bit master to avoid the dither noise from the 24-> 16 conversion.

What do you think ?

That's a 16 bit-ish noisefloor yes.

It's still prudent to upload the 24 bit version instead of 16 bit, though.

Mainly theoretically, since the difference between a dithered 16 bit and original 24 bit source is already extremely low. After lossy conversion I doubt you can even measure the difference, let alone hear it.

The reason isn't to avoid the dither noise from 24 bit > 16 bit dithered per se, but to let the lossy encoding process have as many "original" bits to work with.

Lossy formats like AAC don't have a fixed bit depth and can theoretically benefit from having a 24 bit source.

Thanks a lot Lagerfeldt!




Cheu

Thanks for the information.

Just to clarify. Everything eventually plays at 44.1k on YouTube?

I tend to use mostly 48k, as it seems the best overall compromise. But it's the worst option for YouTube conversion?

I want to know this! i have to set my DAC to 44.1 or 48 to listen to youtube?

Daniel.

If you examine a downloaded video from YouTube and look at the audio stream it's 44.1 kHz regardless of what you uploaded (last I checked, but feel free to do a more recent check).

So playing back at 48 kHz from YouTube would give you 44.1 kHz > realtime aka. low quality SRC > 48 kHz.

Core Audio on Mac handles this in the background and it sounds somewhere between acceptable and crap, especially ruining transient timing with resulting metallic artifacts.

Not sure how it's handled on a Windows PC.

TL;DR: Use 44.1 kHz for playback on YouTube.

Thanks man!, really i don't understand why Youtube ask for 48k audio to then go for a SRC to 44.1k (and then me thinking that the video standard is 48... another SRC)

Best regards,
Daniel.

Good work !
does anyone know the actual youtube playbacks bitrates / codecs ? can't find the very last version they adopted, thanks !

They encode in a variety of codecs and bitrates, and feed a matching one depending on the target device used for playback.

For audio that includes AAC or Opus, all the way down from 24 kbps and up to 192 kbps (for 720p resolution videos and above, on high speed Internet).

thanks a lot! do you have the link of their official page mentioning that? I really could find it. Plus I was sure it was a higher bit rate ate some point (384 or something)

Thanks (again)!

Someone contacted me with information that I feel obliged to add to this thread. I can't confirm it (yet), but it could be relevant:

When/if Opus is used for audio conversion, instead of AAC, then the sample rate is always 48 kHz.

According to the same person Opus is used for 1440p video (and above?).

I hate to muddy the water, but I also hate to not include any potentially relevant information.

If anyone can shed some light on this, be my guest.

My tests above were performed with 1080p HD video and 24 bit WAV and are completely valid for that format. Tested on a computer.

I can't speak for other combinations, i.e. a cellular phone.

youtube-dl will give you information on what audio codecs YouTube is offering:

Note the 3gp targets.

I haven't checked this but I assume everything is encoded in CBR?

Hello.
1. In a page called: "Encoding specifications for music videos" it's written something like that:
"The features described in this article are available only to partners who use YouTube's Content ID matching system"
https://support.google.com/youtube/answer/6039860?hl=en
Does it actually mean, that you can upload wav or loseless, only if you have Content ID? Or it speaks about other features?
Right now it seems to me not logical. If anyway, youtube going to convert this audio file, and reconvert it, if it's already compressed to AAC, then no matter if you have Content ID, or not, this process gonna happen.
2. Is there any updates about sample rate - in the last message @ Lagerfeldt , updated some information about Opus (1440p video (and above?).) So do we already know, if "above", or not. If then, why when we download a video from Youtube, it's always 44100 - seems like double SRC 44k-48k-44k.
Thanks!

No, you can always upload lossless (PCM) audio in your video to YouTube.

And it'll always be converted to a lossy format, yes.

No, and I'm having a hard time reading Johann's readout above your post. In any case we don't know if these numbers were based on a max 1080p video, in which case we're none the wiser.

But I'll keep digging and post back. Perhaps I need to reupload those tests in 1440p.

Ive just been exporting the full 24-96k quality from Final Cut Pro with the ProResHQ setting and uploading that. YouTube should have the same or better SRC we have, and can always make higher quality versions available in the future.

BTW, it’s absurd Youtube still uses 192kbps audio, yet 4K video. Compressed audio starts sounding acceptable at 320kbps, cmon it’s not a big streaming size increase to allow that (like Vimeo does).

With 300 hours of video uploaded every minute, what they use is optimized for speed, rather than quality.

Apologies for being a nobody popping out of nowhere, but as a digital audiophile and enthusiast that's a regular of the Doom9 forums who also regularly uses youtube-dl, I'd like to clear up a few things...


1. All except the lowest of least popular videos and/or super-recent video uploads will receive audio encodes in both AAC and Opus with a bitrate of roughly 128kbps (the Opus encodes tend to commonly hit considerably higher bitrates however).

2. YouTube's AAC audio encodes use a sampling rate of 44.1KHz while their Opus audio encodes use a sampling rate of 48KHz (for one, Opus doesn't even support 44.1KHz)

3. YouTube always defaults to playing with Opus audio is the encodes exist and your browser supports it (AFAIK Safari is the only modern big-name browser that does not support Opus audio)

4. For a given sampling rate and bit rate, Opus should almost if not always be considerably higher quality than AAC

5. You will only ever find 192kbps AAC audio on YouTube'd non-DASH fmt22 encodes on old and/or unpopular videos. In general, if the video also has VP9 video encodes and/or Opus audio encodes, then the AAC audio for the non-DASH fmt22 encode will be at 128kbps (however, I do believe that more recent unpopular videos without VP9 and Opus encodes will still have only 128kbps AAC audio for their non-DASH fmt22 encodes).

6. YouTube supports the MKV video container, so you can very easily combine traditional video formats (AVC, VP9, etc) with a variety of lossless audio codecs (most notably 24bit FLAC) that aren't supported by other common video container formats (e.g. MP4, MOV, WebM, etc).

In my haste I forgot to mention that sometimes (albeit rarely) even less recently-uploaded videos that have received VP9 video encodes can sometimes lack Opus audio encodes.