Which audio quality for compressed video source material?

[Harry_HH]

Lately I've produced few music video clips (about 3-4 minutes long), mainly for the Youtube-kind of use(i.e. compressed video for the use in the Internet).

My question: which audio quality should I use? The file size limit e.g. for the Youtube is 1 GB. For video editing and rendering I'm using the Premiere Pro 2 where the options for the rendering formats are rather limited, best results I've got so far by using the Windows media format, my target video is 1920x1080 (25 fps).

Because I'm a newbie what comes to the video editing, my basic assuption was that because the target is to produce compressed video for the net, the most "economical" suitable audio track is well mastered and maximised Mp3 (256 kbps).

Premiere Pro 2 rendering options includes different kind of audio target qualities, undepending what your audio source material is you can choose diffrent qualites. But if your audio source material is lower quality than the target, I wonder what the result is? The basic assumption is that best result comes when your source quality is as good as possible? But is this really the case (e.g. I've read somewhere that you get better Mp3 from the dithered CD quality than e.g. from the 48 kHz/24 bit material).
For some reason in the Premiere Pro 2 rendering there aren't option for 44 kHz/16 bit/256 kbps (stereo), the nearest good compressed stereo quality is 44 kHz/24 bit/256 kbps. Of course there are options for more compressed audio than 256 kbps but 256 is for my experience compression level which gives a relative good result.

My audio source options in this case are:

1) 48 kHz, 26 bit non dithered maximised (2304 kbps)
2) 44 kHz, 16 bit dithered maximised (1411 kbps)
3) Mp3 generated from the option 2), the quality is 256 kbps

My target quality both for the audio and video you can read above.

Which audio source quality gives the best result in this case?

Harry

[cron]

I always go with uncompressed audio as Youtube transcodes everything. Regardless of the audio you give it, Youtube will always re-encode your audio at:

240p - 64k MP3
360p/480p - 128k AAC
720+ - 192k AAC

I always encode my video with an uncompressed audio track. Just the usual 44.1/16 master with a tiny bit of headroom to allow for encoder clipping. Exactly the same as if I was compressing the audio myself. The size of the audio is so tiny compared to the video that it seems rude not to! I'd rather lose a bit from the image quality than have a 2nd generation transcoded soundtrack if I was really pushed to get under the size limit.

edit: Having said that, we are in the age of laptop speakers and watching YouTube on our phones, so maybe trading audio quality for video quality is the sensible solution, if not the virtuous one.

[schnapsglas]

I would generally mux video and audio together so you can have better control. Really, the best quality should be at well-dithered 16bit loseless with headroom 0.3db, AAC at 256kb+ should give a good result.

[cron]

I would generally mux video and audio together so you can have better control. Really, the best quality should be at well-dithered 16bit loseless with headroom 0.3db, AAC at 256kb+ should give a good result.

If you're going the compressed audio route, might it be worth using a different codec for the audio given that Youtube transcodes to AAC? I recall reading that the worst transcoding artefacts tend to come from transcoding to the same format.

Again, how much difference you'll notice in practice is up for debate. Youtube's audio quality is surprisingly good, and I've little doubt you'll get better audio quality uploading compressed audio to Youtube than you would uploading uncompressed audio to Soundcloud (what with its tiny 128K MP3 streams).

[schnapsglas]

I would generally mux video and audio together so you can have better control. Really, the best quality should be at well-dithered 16bit loseless with headroom 0.3db, AAC at 256kb+ should give a good result.

If you're going the compressed audio route, might it be worth using a different codec for the audio given that Youtube transcodes to AAC? I recall reading that the worst transcoding artefacts tend to come from transcoding to the same format.

Again, how much difference you'll notice in practice is up for debate. Youtube's audio quality is surprisingly good, and I've little doubt you'll get better audio quality uploading compressed audio to Youtube than you would uploading uncompressed audio to Soundcloud (what with its tiny 128K MP3 streams).

Aye, Youtube has much better quality if you allow 480p+, I think. I am not sure about transcoding to same format gives more artifacts -- I have searched Hydrogenaudio and found nothing. I suppose it could be possible if same encoder is used, but I believe at 256kbps+ AAC, it is very close to loseless already. That's my reasoning.**

**If I do ABX, AAC always comes out on the top, and I always found 256kbps to be safe unless I am trying REALLY REALLY hard to look for artifacts. Meaning, I believe loseless -> 256kbps AAC -> loseless -> 256kbps AAC then the first and fourth file are very hard to distinguish. I certainly don't find this to be the case with say, 192kbps MP3 or LAME V2. Also I cannot find the reference, but I think at same rate LAME's lowpass filter is set lower than say Nero's encoder or even iTunes AAC. Only case I know is LAME V0, which doesn't have lowpass. [might be wrong, someone check!]

Youtube's highest audio quality is still 192kbps AAC, however. See: http://en.wikipedia.org/wiki/YouTube#Quality_and_codecs

Oh and yes, keep 44.1khz. That's the right sampling rate.

[Harry_HH]

All most useful points.

A couple of comments/questions: one very fundamental thing was new for me, or I've never given a though for it, namely that Youtube really re-compress your material, despite that they also give a limit to your source material. This, of course, makes the end result quality worse.

From this point seems that there are no reason to compress you audio when you render your video. But what if there's not an option "no audio compression" in you video editor?

An other point: my usual headroom setting in mastering maximiser is 0,1 db - now I should listen carefully the Youtube end result with this setting, should I re-master the file with the 0,3 db headroom to avoid distortion/artifacts etc.

H.

[schnapsglas]

From this point seems that there are no reason to compress you audio when you render your video. But what if there's not an option "no audio compression" in you video editor?

Most video editors do have PCM output, even windows movie maker, I think (???) But there is a reason. First:

1. For storage, it makes no sense to have an encoded video and loseless audio together;
2. If it is to be distributed, it makes sense to have the encoding done in case it needs to be sent to other places;
3. Really, there is a big difference. PCM at 16/44.1 is really large. You can get by with quarter of that bandwidth with good codecs. I mean, no one would stream PCM 16/44.1. It just doesn't make sense for me.

If I really have to archive, I just use AAC and call it a day. If it is something that requires me to have 4:2:2 video, then I do the sensible thing and just take out the audio and FLAC everything. Saves space in the end, and they can be recalled quite easily.

Depends on your limiter, but I never go above -.2db.