Intersample Overs, and why your mp3 sounds bad |
|
|
|
|
Dec 1 2011, 04:33 PM
|
|
This thread is partly in response to a question I was asked recently on our facebook page. It does however seem timely as we've had a number of audio mixes sent to us recently that suffer from this issue.
Now many people, including a lot of audio engineers and daw manufacturers will tell you that digital 0 (commonly 0dbFS) is the ceiling for digital audio and that you can't produce a digital audio signal higher than 0dBFS. They also will show you nice waveforms where a peak is flat topped at 0dBFS. That is only sort of true. Sort of because it is possible for a signal to exceed 0dBFS and this is why. First it is necessary to remember that digital audio is just discrete binary code - a lot of zeros and ones. Analogue audio - the sort that you hear - however is a continuous signal or wave and is not discrete. A digital signal consists of numbers 0s and 1s. For us to hear it as audio it has to be converted into an audible signal and this is made possible because the DAC (digital to audio convertor) changes those 0s and 1s to a voltage signal that is then sent to an amplifier and monitor: the DAC has to interpret binary code and translate it to continuous voltage. For a DAC to do this it first has to convert the code to discrete separate voltage impuses. It then has to add extra impulses (called interpolation) between the real impulses so that a (near)continuous stream can be produced by a reconstruction filter. Now a DAC has to interpolate what happens between two real impulses in order to decide how to construct it's extra impulses to go between the real ones. If the real ones are increasing in intensity then the dac will interpolate this to mean that it must add extra impulses that also continue this series. Most of the time in digital to analogue conversion there is enough data for the interpolation to be reasonably accurate. However there is an instance when it can and does go wrong and this is the intersample over. If you have a set of very fast transient peaks that are very close to or at 0dBFS you would assume that they would not exceed 0dBFS. After all everyone tells us that 0dBFS is the ceiling for digital. You would be wrong. A series of fast transients can go over because of intersampling overs. Here the dac reconstructs the transients but due to summing error some of the previous peak's impulse value can be added in to the next peak. (I'm simplifying a lot here by the way and there are related issues concerned with aliasing in the reconstruction.) So if a series of fast transient are close to 0dBFS then summing in the series can result in several of them exceeding 0dBFS. Consumer and prosumer DACs will not play any signal that exceeds 0dBFS . So if you have transients that intersample over the result will not be nice crisp drum transients but distortion. You can end up with a sound that sounds particularly nasty on small speakers. Now fast drum transients are not the only way to get intersample overs - hard limiting and hard clipping are two other ways. If you look at a goniometer spectra for audio that has been hard brickwalled limited the signal will often approximate a diamond or similar rhombus shape rather than the more circular/oval shape of a stereo feed. Transcoding a wave to an mp3 can result in internal summing as well due to the mp3 encoding. This can lead to IS overs as well even if the original wave file is not clipped but only approaches 0dBFS. I've seen a lot of these IS overs in mp3 transcoding in the last few weeks and it's becoming more the norm as people are moving to digital distribution for audio rather than using CD. So what can you do about it? First it is imporant to keep some distance between your peak signal and 0dBFS. This is particularly true if you intend to transcode the file to mp3. Some mastering engineers recomend a value of -0.3dBFS, I find however that that works for some material but not all. For a lot of heavily compressed material we peak lower at anywhere between -0.5 down to -1dBFS if we have to transcode and -0.3 if we don't. Second you should monitor for it. Be aware though that the peak meter in your daw is probably not very accurate and does not measure signal and so will not do this. In the first instance most daw peak meters only show 1 in 4 of the samples that it measures. So 3 samples could clip before the meter will show a clip. Second you need to remember that the meter in a daw relates to sample not signal - that is these meters do not register and show intersample over signals. To be able to read IS overs a meter needs to oversample and accurately reconstruct the signal and very few vst meters can do this. We routinely check both the mixes that we are sent and our final masters for ISOs as we have meters capable of doing this. The growing trend for mixes to have ISos lead me to believe that very few people do this, or are capable of doing it, though. Intersample overs are enough of an issue in mastering that the EBU (European Broadcast Union - the guys who set the standards for all video and audio broadcast in Europe) have now issued a new standard (ITU-R BS1770-2 standard) for oversampling meters and we may now use a new level scale called dBTP alongside the usual odBFS. There is also some push for us all to adopt the new EBU R128 directive that relates to how we deifne and measure loudness. For people who work in video and film the EBU R128 directive could have an immediate impact but it may take more time for it to filter through to audio. (Rob you probably really need to be aware of these new standards for your Cruciform work) -------------------- Get your music professionally mastered by anl AES registered Mastering Engineer. Contact me for Audio Mastering Services and Advice and visit our website www.miromastering.com
Be friends on facebook with us here. We use professional, mastering grade hardware in our mastering studo. Our hardware includes: Cranesong Avocet II Monitor Controller, Dangerous Music Liasion Insert Hardware Router, ATC SCM Pro Monitors, Lavry Black DA11, Prism Orpheus ADC/DAC, Gyratec Gyraf XIV Parallel Passive Mastering EQ, Great River MAQ 2NV Mastering EQ, Kush Clariphonic Parallel EQ Shelf, Maselec MLA-2 Mastering Compressor, API 2500 Mastering Compressor, Eventide Eclipse Reverb/Echo. |
 |
 
|
|
|
|
|
|
Dec 1 2011, 06:13 PM
|
|
Very interesting article you wrote tony. Never thought about it. Can this be a rule that we should all follow when mastering our recordings? I'm doing a mastering directly on the master bus, not sure if this is the good way of doing it tho.. Should I use some metering (except the regular stereo channel meter) before the signal goes out?
While we are on the topic, I have a question regarding encoding audio from vegas. I noticed some of big companies ask that projects should be rendered with at least -5dB. I noticed that some tracks that are compressed, and go beyond that (let's say peak is @ -1dB), the rendered file has some clipping here and there. Is this due to this matter or matter of codec? -------------------- - Ivan's Video Chat Lesson Notes HERE
- Check out my GMC Profile and Lessons - (Please subscribe to my) YouTube Official Channel - Let's be connected through ! Facebook! :) |
|
|
|
|
|
|
|
|
Dec 2 2011, 02:03 AM
|
|
Hey Tony,
Thank you for that indepth read. I'll be back to digest it when I can. I really appreciate and learn from all your commentary.  -------------------- |
|
|
|
|
|
|
|
|
|
Dec 2 2011, 10:02 AM
|
|
QUOTE (Ivan Milenkovic @ Dec 1 2011, 05:13 PM)  Very interesting article you wrote tony. Never thought about it. Can this be a rule that we should all follow when mastering our recordings? I'm doing a mastering directly on the master bus, not sure if this is the good way of doing it tho.. Should I use some metering (except the regular stereo channel meter) before the signal goes out? While we are on the topic, I have a question regarding encoding audio from vegas. I noticed some of big companies ask that projects should be rendered with at least -5dB. I noticed that some tracks that are compressed, and go beyond that (let's say peak is @ -1dB), the rendered file has some clipping here and there. Is this due to this matter or matter of codec? Should you use some extra metering - yes, though don't become obsessed with watching meters as critical listening is much more important than watching a meter. If you can then use an oversampling meter or a hardware vu rather than just the digital peak sample meters that are more common in recording daws. If you can't then see if you can at least reset the daws meters so that they read and display every sample rather than every 3rd or 4th. I've never used Vegas so can't say for certain Ivan but it sounds to me like it's both. It's both as the codec is summing the transients on the compressed audio and that produces the increased signal, which leads to IS overs and clipping. I'd try using less compression and be conservative with your levels if you have to use that codec. There are other possible reasons though. To sum a peak from -1dB to clipped is pretty extreme, most summing gain of this sort is usually in the order of .1dB up to .5dB. So normally output limiting at -.3dB before transcoding is sufficient. If you're getting clipping at -1dB on transcoding I kind of wonder if you also need to check your gainstaging. You may well be clipping going in to the ADC without realising it and/or some of your vst chain also may not be gainstaged properly, particularly if some are fixed rather than floating point. BTW I'm assuming that the companies are asking for a peak at -5dB as the intended final destination is broadcast film/video. (Sounds kind of like EBU standards here.) If that is the case it's often a good idea to minimise the amount of compression and limiting you would normally use and above all else do not clip. (When I used to work in broadcast audio we routinely rejected any audio file that was clipped. A major reason was that the additional processing and transcoding that we did would make a clipped file cascade: if you're lucky you get thin, weak audio; if you're unlucky you get a lot of nasty continuous white noise.)That really does mean that you need to be more conservative with your levels and that your recordings have proper crest and peak well below the 0dbFS. It also means that you should gainstage everything feeding your ADC and also coming out to the DAC. If you're doing a lot of post work here it may well be worth investing in decent hardware Durroughs meters and getting your monitor chain fully calibrated and set up. QUOTE (OzRob @ Dec 2 2011, 01:03 AM) Hey Tony, Thank you for that indepth read. I'll be back to digest it when I can. I really appreciate and learn from all your commentary. NP Rob and glad to help. It's possiibly not an easy read as it gets bit technical very quickly even though I'm simplifying a lot. -------------------- Get your music professionally mastered by anl AES registered Mastering Engineer. Contact me for Audio Mastering Services and Advice and visit our website www.miromastering.com
Be friends on facebook with us here. We use professional, mastering grade hardware in our mastering studo. Our hardware includes: Cranesong Avocet II Monitor Controller, Dangerous Music Liasion Insert Hardware Router, ATC SCM Pro Monitors, Lavry Black DA11, Prism Orpheus ADC/DAC, Gyratec Gyraf XIV Parallel Passive Mastering EQ, Great River MAQ 2NV Mastering EQ, Kush Clariphonic Parallel EQ Shelf, Maselec MLA-2 Mastering Compressor, API 2500 Mastering Compressor, Eventide Eclipse Reverb/Echo. |
|
|
|
|
|
|
|
|
|
Dec 3 2011, 06:37 PM
|
|
Excellent answer, thanks Tony!
-------------------- - Ivan's Video Chat Lesson Notes HERE
- Check out my GMC Profile and Lessons - (Please subscribe to my) YouTube Official Channel - Let's be connected through ! Facebook! :) |
|
|
|
|
|
|
|
|
Dec 3 2011, 07:11 PM
|
|
fascinating ... Thanks Tony!
-------------------- Check out my Instructor profile
Live long and prosper ... My Stuff: Electric Guitars : Ibanez Jem7v, Line6 Variax 700, Fender Plus Strat with 57/62 Pickups, Line6 Variax 705 Bass Acoustic Guitars : Taylor 816ce, Martin D-15, Line6 Variax Acoustic 300 Nylon Effects : Line6 Helix, Keeley Modded Boss DS1, Keeley Modded Boss BD2, Keeley 4 knob compressor, Keeley OxBlood Amps : Epiphone Valve Jnr & Head, Cockburn A.C.1, Cockburn A.C.2, Blackstar Club 50 Head & 4x12 Cab |
|
|
|
|
|
|
|
|
|
Dec 6 2011, 01:51 PM
|
|
Hey Tony,
I note the new Flux Elixir complies with the standard you mention. Have you tried it out yet? (Not for that reason, but just in general.) -------------------- |
|
|
|
|
|
|
|
|
|
Dec 6 2011, 04:32 PM
|
|
QUOTE (OzRob @ Dec 6 2011, 12:51 PM) Hey Tony, I note the new Flux Elixir complies with the standard you mention. Have you tried it out yet? (Not for that reason, but just in general.) Hadn't realised that so thanks for the heads up Rob . Don't really need it though as we've already have stuff that complies .
-------------------- Get your music professionally mastered by anl AES registered Mastering Engineer. Contact me for Audio Mastering Services and Advice and visit our website www.miromastering.com
Be friends on facebook with us here. We use professional, mastering grade hardware in our mastering studo. Our hardware includes: Cranesong Avocet II Monitor Controller, Dangerous Music Liasion Insert Hardware Router, ATC SCM Pro Monitors, Lavry Black DA11, Prism Orpheus ADC/DAC, Gyratec Gyraf XIV Parallel Passive Mastering EQ, Great River MAQ 2NV Mastering EQ, Kush Clariphonic Parallel EQ Shelf, Maselec MLA-2 Mastering Compressor, API 2500 Mastering Compressor, Eventide Eclipse Reverb/Echo. |
|
|
|
|
|
|
|
|
Dec 6 2011, 05:47 PM
|
|
Wow - I actually understood this post
 Great write up Tony, as always! Now I have to re-mix everything I've ever done :/ (jk). Chris! -------------------- The more I practice, the more I wish I had time to practice!
My Band Forum: http://passionfly.site/chat  |
|
|
|
|
|
|
|
|
|
Dec 7 2011, 11:26 AM
|
|
QUOTE (Sinisa Cekic @ Dec 6 2011, 09:41 PM) ... What can you tell me about FLAC format. Many people say that is much better than Mp3, and yet I believe that many of them can't hear the difference between them, me too. Does the file size is only important to Mp3 assumes primacy as the most widespread audio format ?! FLAC is an audio lossless compression codec whilst mp3 is a lossy compression format. In the case of FLAC it can reduce the size of an audio file by around 50% (so there are more efficient compression codecs than FLAC). It only works on fixed point samples: the codec uses integer maths so there is less likely/no need to resample for interpolation. That's fine for PCM encoded audio but it may be worth noting that some daws are floating point all the way prior to final rendering. FLAC can also generate and read a cue sheet so it can be set to properly replicate the PQ of an audio CD and so also produce the correct track placements and gaps. It's used as the EBU standard for transmitting audio in EUrope between audio broadcast stations. Is it better than mp3? Audio wise yes because it is lossless. Listen in particular to music that has a wide dynamic range and that was originally well recorded and produced before transcoding to mp3. With mp3s - even those that have been properly encoded - you can very often hear a loss of detail: in particular the quieter instruments seem to recede and often dissappear. This is partly because the mp3 codec raises the noise floor of the audio significantly so any low level detail gets lost. With mp3s you can also hear spurious unwanted random noise if you have periods of silence in the audio (it's there in the non-silent sections but tends to be covered up/masked by all the music). With poorly encoded mp3s you also can often hear distortion that is particularly noticeable in the high end frequencies especially if you listen to the mp3 on headphones. Here you often hear the distortion setting in on the left channel some time before the right (or vie versa). So with mp3s you have a poor noise floor which results in the loss of detail and quiet sounds, spurious unwanted nose and potenital distortion. The sound also tends to sound thin and edgy. FLAC - assuming the material was original well recorded and produced - shouldn't do any of that. However a 50% compression of a 50 meg audio file (around about a 4 minute 16/44.1) would result in something between 20-25 meg in size. If you went to a 128 vbr mp3 (not great quality but one that is used for audio srteaming a lot) the file size would be about 4 1/4 meg and at 320 cbr (best quality for an mp3) about 10 meg. With FLAC you must also remember that a lot of domestic players don't have a FLAC codec as standard and so can't play them. Sorry Sinisa - meant to add this bit and forgot... Is FLAC better for audio then mp3 - IMHO yes. Will it take over from mp3? IMHO no because mp3 (along with AAC) is already seen as the preferred consumer format and so has a massive foothold/stranglehold on the market. It's a bit like with old videos - Betamax was better than VHS quality wise but VHS won as it dominated the market. The issue here to me is that there are literally 100s of millions (possibly billions) of personal players and consumer hifis that play mp3 'out of the box' but that do not come with FLAC as standard. One other thing to add is that all those who advertise their 320 CBR mp3s as 'high quality CD like audio downloads' clearly know very little about audio quality  .
-------------------- Get your music professionally mastered by anl AES registered Mastering Engineer. Contact me for Audio Mastering Services and Advice and visit our website www.miromastering.com
Be friends on facebook with us here. We use professional, mastering grade hardware in our mastering studo. Our hardware includes: Cranesong Avocet II Monitor Controller, Dangerous Music Liasion Insert Hardware Router, ATC SCM Pro Monitors, Lavry Black DA11, Prism Orpheus ADC/DAC, Gyratec Gyraf XIV Parallel Passive Mastering EQ, Great River MAQ 2NV Mastering EQ, Kush Clariphonic Parallel EQ Shelf, Maselec MLA-2 Mastering Compressor, API 2500 Mastering Compressor, Eventide Eclipse Reverb/Echo. |
|
|
|
|
|
|
|
|
|
Dec 11 2011, 11:08 PM
|
|
Another marketing trick, and interesting point about FLAC in general. Great answer!
-------------------- - Ivan's Video Chat Lesson Notes HERE
- Check out my GMC Profile and Lessons - (Please subscribe to my) YouTube Official Channel - Let's be connected through ! Facebook! :) |
|
|
|
|
|
|
|
|
Dec 11 2011, 11:37 PM
|
|
Great explanation ,thanks Tony
|
|
|
|
|
 
|
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:
