I mix and master in Ableton, but the .WAV then goes into Premiere Pro and is bounced as an H.264 (in the video) with its ultimate destination being YOUTUBE. So the .WAV from Ableton goes into PP and is then exported as AAC.
I've been experimenting with the best way to do this for highest audio quality. I'm big fan of 320 KBPS .WAV's and am trying to encapsulate that quality as it finally ends on YouTube.
Does anyone have experience or tips with this? I'm doing pretty good, but just stumbled across this forum with its own AAC section and figured there may be some experts.
- Is there a specific format that AAC likes better, besides .WAV? I can bounce .AIFF's or FLAC's from Ableton.
- Is it smart to Dither if bouncing from the DAW (to .WAV)? Or should Dithering be done in Premiere Pro after the .WAV is out ... before its converted to AAC.
- I am exporting at a 24 Bit Depth from Ableton with a 48,000 sample rate. Does AAC play best with this?
- I know some .MP3's shave off information more so than .WAV's, so you want to be careful pushing 0 dBSF on the limiter so you don't lose that information if the ultimate destination is an .MP3. Is this the same with .AAC?
- Does anyone know if youtube is altering/converting/encoding the .AAC file, or does it leave it as is?
- I am currently bouncing the .AAC's using the "5.1" channel setting, because it allows me to export at 640 kbps, which I do because I assume Youtube will compress it. Is this ideal?
I think that's it for now. These things fascinate me. Sorry if this is ... a bore.
Here's an example of an .AAC file, before it goes into YouTube. This ones actually at 320 kbps but probably not noticeable to most unless you're a golden retriever.
https://www.dropbox.com/s/e0k1fc6s96eaxrf/Louder%203.aac?dl=0 (https://www.dropbox.com/s/e0k1fc6s96eaxrf/Louder%203.aac?dl=0)
If you have the upload bandwidth, then upload stuff like cineform or prores (video) + uncompressed audio (wav), afaik youtube will always re-compress everything to many lossy codecs. You can analyze what it may return with a tool like youtube-dl:
youtube-dl -F "https://www.youtube.com/watch?v=QGzxkRileIM"
[youtube] QGzxkRileIM: Downloading webpage
[youtube] QGzxkRileIM: Downloading video info webpage
[info] Available formats for QGzxkRileIM:
format code extension resolution note
249 webm audio only DASH audio 60k , opus @ 50k, 7.45MiB
250 webm audio only DASH audio 77k , opus @ 70k, 9.58MiB
140 m4a audio only DASH audio 129k , m4a_dash container, mp4a.40.2@128k, 20.01MiB
171 webm audio only DASH audio 138k , vorbis@128k, 16.24MiB
251 webm audio only DASH audio 149k , opus @160k, 18.92MiB
278 webm 256x144 144p 97k , webm container, vp9, 30fps, video only, 12.53MiB
160 mp4 256x144 144p 111k , avc1.4d400c, 30fps, video only, 7.09MiB
242 webm 426x240 240p 224k , vp9, 30fps, video only, 21.75MiB
133 mp4 426x240 240p 246k , avc1.4d4015, 30fps, video only, 13.12MiB
243 webm 640x360 360p 420k , vp9, 30fps, video only, 41.92MiB
134 mp4 640x360 360p 633k , avc1.4d401e, 30fps, video only, 33.02MiB
17 3gp 176x144 small , mp4v.20.3, mp4a.40.2@ 24k, 12.34MiB
36 3gp 320x180 small , mp4v.20.3, mp4a.40.2, 33.43MiB
18 mp4 640x360 medium , avc1.42001E, mp4a.40.2@ 96k, 70.03MiB
43 webm 640x360 medium , vp8.0, vorbis@128k, 104.56MiB (best)
If you have bandwidth, I would use flac, dither 24 bit down to 16 bit, and make sure when you go from float to int that you aren't clipped.
I've been experimenting with the best way to do this for highest audio quality. I'm big fan of 320 KBPS .WAV's and am trying to encapsulate that quality as it finally ends on YouTube.
320kbps is NOT uncompressed WAV. "CD quality" 16-bit, 44.1kHz, 2-channel WAV files have a bitrate of 1411kbps. (16 x 44,100 x 2). 320kbps is probably MP3, possibly in a WAV container.
- Is there a specific format that AAC likes better, besides .WAV? I can bounce .AIFF's or FLAC's from Ableton.
When you're using lossy compression it's best to use a lossless format and compress once as the final step.
- Is it smart to Dither if bouncing from the DAW (to .WAV)? Or should Dithering be done in Premiere Pro after the .WAV is out ... before its converted to AAC.
The "rule" is, dither when downsampling the bit-depth. But, at 16-bits or better you can't hear dither or the effects of dither. You shouldn't dither when converting to a lossy format.
You shouldn't dither when compressing losslessy either, unless you are downsampling at the same time.
- I know some .MP3's shave off information more so than .WAV's, so you want to be careful pushing 0 dBSF on the limiter so you don't lose that information if the ultimate destination is an .MP3. Is this the same with .AAC?
I believe it's the same with AAC. With lossy compression the wave shape changes. Some peaks get a little higher and some peaks get a little lower. If you normalize to 0dB, some peaks may go over 0dB after (file) compression. It tends to be worse if the original is (dynamically) compressed/limited.
Note that the MP3 isn't clipped... It can go over 0dB without clipping. The DAC can clip if you play-back at full-digital volume but as far as I know the slight clipping caused by lossy compression is not audible... If you hear a compression artifact it's probably something else. But, some people do normalize to -1dB or so before MP3 compression.
I've been experimenting with the best way to do this for highest audio quality. I'm big fan of 320 KBPS .WAV's and am trying to encapsulate that quality as it finally ends on YouTube.
320kbps is NOT uncompressed WAV. "CD quality" 16-bit, 44.1kHz, 2-channel WAV files have a bitrate of 1411kbps. (16 x 44,100 x 2). 320kbps is probably MP3, possibly in a WAV container.
My mistake. The .WAV's I bounce from Ableton are 24 bit 48,000.
They aren't 320 KBPS anything. Live 10 now bounces a 320 KBPS .MP3 file with the .WAV so I think I just associated that in my head. Thanks for the correction.
Thanks Doug.
Do you bounce any projects to Youtube? What's your preferred process/format?
I'm considering trying ProRes or Cineform at the suggestion of the above replies.
I've been bouncing to H.264 4K because that's what YouTube says it likes, but I have no upload bandwith constraints so absolute quality is really the priority.
If you have bandwidth, I would use flac, dither 24 bit down to 16 bit, and make sure when you go from float to int that you aren't clipped.
I'll try out FLAC and see how it sounds and dither down to 16 bit.
I'm not versed enough in video encoding to understand the float to int recommendation I'll have to do some research on that.
Thanks,
If you have the upload bandwidth, then upload stuff like cineform or prores (video) + uncompressed audio (wav), afaik youtube will always re-compress everything to many lossy codecs. You can analyze what it may return with a tool like youtube-dl:
youtube-dl -F "https://www.youtube.com/watch?v=QGzxkRileIM"
[youtube] QGzxkRileIM: Downloading webpage
[youtube] QGzxkRileIM: Downloading video info webpage
[info] Available formats for QGzxkRileIM:
format code extension resolution note
249 webm audio only DASH audio 60k , opus @ 50k, 7.45MiB
250 webm audio only DASH audio 77k , opus @ 70k, 9.58MiB
140 m4a audio only DASH audio 129k , m4a_dash container, mp4a.40.2@128k, 20.01MiB
171 webm audio only DASH audio 138k , vorbis@128k, 16.24MiB
251 webm audio only DASH audio 149k , opus @160k, 18.92MiB
278 webm 256x144 144p 97k , webm container, vp9, 30fps, video only, 12.53MiB
160 mp4 256x144 144p 111k , avc1.4d400c, 30fps, video only, 7.09MiB
242 webm 426x240 240p 224k , vp9, 30fps, video only, 21.75MiB
133 mp4 426x240 240p 246k , avc1.4d4015, 30fps, video only, 13.12MiB
243 webm 640x360 360p 420k , vp9, 30fps, video only, 41.92MiB
134 mp4 640x360 360p 633k , avc1.4d401e, 30fps, video only, 33.02MiB
17 3gp 176x144 small , mp4v.20.3, mp4a.40.2@ 24k, 12.34MiB
36 3gp 320x180 small , mp4v.20.3, mp4a.40.2, 33.43MiB
18 mp4 640x360 medium , avc1.42001E, mp4a.40.2@ 96k, 70.03MiB
43 webm 640x360 medium , vp8.0, vorbis@128k, 104.56MiB (best)
'
Great idea! I got this up and running here's what it says about my video.
[youtube] G8HrgC-9Gc0: Downloading webpage
[youtube] G8HrgC-9Gc0: Downloading video info webpage
[info] Available formats for G8HrgC-9Gc0:
format code extension resolution note
249 webm audio only DASH audio 56k , opus @ 50k, 2.28MiB
250 webm audio only DASH audio 76k , opus @ 70k, 3.02MiB
140 m4a audio only DASH audio 128k , m4a_dash container, mp4a.40.2@128k, 6.14MiB
171 webm audio only DASH audio 149k , vorbis@128k, 5.67MiB
251 webm audio only DASH audio 150k , opus @160k, 5.97MiB
278 webm 256x144 144p 112k , webm container, vp9, 30fps, video only, 3.92MiB
160 mp4 256x144 144p 128k , avc1.4d400c, 30fps, video only, 3.80MiB
133 mp4 426x240 240p 285k , avc1.4d4015, 30fps, video only, 8.74MiB
242 webm 426x240 240p 334k , vp9, 30fps, video only, 7.93MiB
134 mp4 640x360 360p 735k , avc1.4d401e, 30fps, video only, 20.44MiB
243 webm 640x360 360p 1193k , vp9, 30fps, video only, 18.31MiB
135 mp4 854x480 480p 1352k , avc1.4d401f, 30fps, video only, 39.78MiB
244 webm 854x480 480p 2296k , vp9, 30fps, video only, 34.73MiB
136 mp4 1280x720 720p 2717k , avc1.4d401f, 30fps, video only, 80.63MiB
137 mp4 1920x1080 1080p 4940k , avc1.640028, 30fps, video only, 165.10MiB
247 webm 1280x720 720p 6548k , vp9, 30fps, video only, 86.72MiB
248 webm 1920x1080 1080p 12699k , vp9, 30fps, video only, 164.35MiB
271 webm 2560x1440 1440p 21436k , vp9, 30fps, video only, 373.19MiB
313 webm 3840x2160 2160p 37368k , vp9, 30fps, video only, 793.82MiB
17 3gp 176x144 small , mp4v.20.3, mp4a.40.2@ 24k, 3.43MiB
36 3gp 320x180 small , mp4v.20.3, mp4a.40.2, 10.66MiB
18 mp4 640x360 medium , avc1.42001E, mp4a.40.2@ 96k, 33.81MiB
43 webm 640x360 medium , vp8.0, vorbis@128k, 57.66MiB
22 mp4 1280x720 hd720 , avc1.64001F, mp4a.40.2@192k (best)
Thanks for the ProRes and Cineform suggestions, I'll give those a try.
My experience in upload to youtube shows me that if you use V1 HE-AAC, or >190 LE-AAC, for example, youtube will convert to 120k LE-AAC and its crops the sound above 16kHz :( :( :( :(.
The best result I observed in streams sites was upload mp3lame VBR (0) audio streams at 48100 Hz encoded inside the H.264 video, them the server choose the better compress method, like done in .wav files.
Some youtube accounts appear have special permissions that allows the 190k AAC ( mp4a.40.2@192k), but its apear to be automatic based in popularity (like the VEVO i.e.). The mp3lame VBR (0) at least "brings" the sound until 20 kHz to youtube server, i think, in "sane" internet bandwidth .
HE-AAC appear to have problems in youtube because the codec licences related at this format.
LE-AAC are 99% converted to poor 120k.
youtube accepts mkv files with embedded flac (24 bit is fine)
uploading 24bit pcm or 24bit flac gives the best audio quality
on upload stereo 256kbps aac-lc is the bare minimum
384kbps aac-lc is reccomended for 2 channel
upload as 2 channel if the content is 2 channel
youtube should take any aac file supported by the specification
529 at 44.1 / 576 at 48 / 1152 at 96 (supported by the standard but idk if youtube will accept this)
regardless for best quality upload 24bit pcm or wav or flac and target -12 LUFS to prevent youtube from normalising on playback
make sure you upload at native sample rate (44.1 or 48khz or whatever)
youtube always resamples depending on whether playback is h264 (44.1khz) or not (48khz)
for most users the correct audio setting in windows for online is 48khz
I have probably thought this before, but: Youtube has changed formats before, and so I take it that they store the original upload for you? Might it be worth something to have it uploaded in a format you want to have backed-up?
I know that it is unsafe as backup, given how Youtube has supported copyright trolls like Universal Music Group when they try to snatch artists' rights (https://www.techdirt.com/articles/20150317/10513630347/umg-licenses-indie-artists-track-then-uses-content-id-to-claim-ownership-it.shtml), but it is hardly worse than nothing?
The best seems to upload whatever container with FLAC in there. as Matroska supports FLAC, you can use that, no problem.
Using FLAC in MKVs also circumvented the 8 channels limitation of the FLAC container for me. Youtube seems to handle it well enough. I don't think Youtube keeps your original, uploaded file. However, every time Youtube changes something with their encoding chain, it might be a good idea to replace the video, i.e. re-upload it. Eventually, Youtube will serve lossless audio, I'm sure. Probably in not too much time, either.
why would anyone upload more then 8 channels of audio to youtube?
please explain
also just in case anyone is confused matroska is mkv
finally never dither down before uploading lossless or wav to youtube
native bitrate and khz is preferred
exceptions are dsd, 32bit or higher, higher then 196khz (not 100% sure)
youtube accepts mkv files with embedded flac (24 bit is fine)
uploading 24bit pcm or 24bit flac gives the best audio quality
on upload stereo 256kbps aac-lc is the bare minimum
384kbps aac-lc is reccomended for 2 channel
Not sure why you bumped an old thread to say this, but no, don't upload 24 bit audio to youtube, dither it down to 16 bit (if FLAC). If lossy it doesn't matter, the codec will handle that for you. You're just wasting time otherwise.
finally never dither down before uploading lossless or wav to youtube
You should always dither down if you're operating above 16 bit and lossless. It probably doesn't matter for lossy.
native bitrate and khz is preferred
exceptions are dsd, 32bit or higher, higher then 196khz (not 100% sure)
Native sampling rate (not bitrate) is preferred if it is 48kHz. Otherwise, you should probably resample, except possibly for 44.1k. Native bitrate is never preferred.
Both native sample rate (as long as it is 44.1 or higher) and native bitrate (16 or 24bit) is now preferred.
Youtube re-encodes everything to multiple sampling rates. There preference for 24bit files is similar to apple's in mastered for itunes (targeting 16 bit aac). Youtube is also now normalizing audio user side but may be re-encoding files and adjusting gain (consistently over the full video when 24 bit files are provided
If you feel dithering 1st provides higher quality then do an AB test. But youtube wants videos to be uploaded once and reprocessed by them as necessary
I believe that youtube is internally testing something potentially along the same lines as MQA.
writing on the wall
https://support.google.com/youtube/answer/6039860?hl=en
https://support.google.com/youtube/answer/1722171?hl=en (expand bitrate and look at the audio bitrate!!! suggestion 384kbps aac-lc - this coincidentally is higher then the 256 min and 320 recommended (although 24 bit pcm or flac preferred) on the other page.
384kbps (stereo) just happens to be the lowest quality where it makes sense to bother with 96khz)
Why is dithering prior to upload better?
Youtube re-encodes everything to multiple sampling rates. There preference for 24bit files is similar to apple's in mastered for itunes (targeting 16 bit aac). Youtube is also now normalizing audio user side but may be re-encoding files and adjusting gain (consistently over the full video when 24 bit files are provided
Youtube's guidelines assume that files will be prepared incorrectly and encourage people to not try and do it themselves. That doesn't mean you can't do a better job yourself if you know what you're doing. If you understand what you're doing, you can set the levels correctly yourself, dither down to 16 bit FLAC, and produce perfectly fine output.
If you feel dithering 1st provides higher quality then do an AB test. But youtube wants videos to be uploaded once and reprocessed by them as necessary
You should always dither when reducing sample size. An AB test can tell you if the improvement you get is audible, which is a different question.
384kbps (stereo) just happens to be the lowest quality where it makes sense to bother with 96khz)
It never makes sense to use 96khz.
Why is dithering prior to upload better?
Dithering down to 16 bit when using lossless saves bandwidth, and something similar (and possibly worse) will ultimately be performed anyway on their end. The advantage of doing it yourself is that you can make sure it is done properly and listen to the final output, and then save bandwidth uploading. If you're going to upload AAC files, none of this matters as the codec will be requantizing everything anyway.
it makes sense to upload in whatever native sample rate you have available.
if you are working with 96khz. upload 96khz.
the more sample rate changes (in the digital domain) the more aliasing and noise.
you are making a good case for working in 48khz (or 44.1khz). It wasn't so long ago that engineers were da/ad'ing sample rate changes because the results were more predictable.
to be perfectly honest i cannot think of a good reason for anyone thats not a professional or has a specific need to work in anything higher then 48khz
if cd's matter then work in 44.1khz
if they don't work in 48khz
it makes sense to upload in whatever native sample rate you have available.
if you are working with 96khz. upload 96khz.
the more sample rate changes (in the digital domain) the more aliasing and noise.
Everything will eventually be converted to 48 KHz, so if you must work at higher sampling rates for some reason, you should convert to 48k at the end. With proper choice of software there will be negligible aliasing and no noise.
to be perfectly honest i cannot think of a good reason for anyone thats not a professional or has a specific need to work in anything higher then 48khz
96k+ is mostly clueless people. From that point of view, it makes sense for Google to recommend not resampling down - people using odd sampling rates are more likely to be uninformed and do the wrong thing.
why would anyone upload more then 8 channels of audio to youtube?
please explain
Because why not? 22.2 or Dolby Atmos support a large number of channels to create a more immersive experience. Youtube is no exception to that, or at least shouldn't be. I'm sure Google/Alphabet is right on top of that game, given that you can rent movies on Youtube, etc. If anything, I'm sure it's in their interest to open up Youtube for something like Netflix and/or Amazon Prime. I don't see why regular users would be barred from using large number of channels, or object oriented audio, as it is with Atmos.
also just in case anyone is confused matroska is mkv
Yeah, I don't think anyone is confused by that ¦D
finally never dither down before uploading lossless or wav to youtube
native bitrate and khz is preferred
Dither
down? It seems you're misunderstanding the way dithering works. Dithering is a form of better reconstructing the resolution of a signal, be it one or two dimensional. Do you mean downsample, perhaps? Dithering is used to better represent a signal at lower bit depths. I'm not quite sure I'm understanding the "dithering down" sense right here.
exceptions are dsd, 32bit or higher, higher then 196khz (not 100% sure)
I don't understand that statement.
dsd bitstream uploads are experimental and/or unsupported by youtube
higher then 24 bit (example 32bit) audio are experimental and/or unsupported by youtube
higher then 192khz sampling rate (example 384khz) are experimental and/or unsupported by youtube
24 bit is dithered down when converted to 16 bit
everything uploaded to youtube is available in both 44.1khz and 48khz
the preference for native sampling rates (higher is better) and full 24 bit audio likely has to do with intent or testing or implementation of MQA or similar technologies. if MQA was open source youtube would already have some sort of public implementation (even if beta).