Sample Rate conversion

Discussion in 'Converters / Interfaces' started by jscott, Apr 4, 2002.

  1. jscott

    jscott Guest

    I've been using Logic for sometime now. However, I've not really attempted to put anything on to a CD yet.

    I don't want this to turn into a debate about sample rates, etc., but I have been using 48 as opposed to 44.1. And I combine midi tracks with Audio. To get the basis for the .wav file, I go out and bring the midi tracks back in as an audio file, and then do the "Bouce" at 16 bit, which means the .wav file is now 16/48.

    So the question is, what method have you found best to convert down to 16/44.1 and retain quality for a CD. It would appear the only way to do this in Logic is "The Digital Factory" in the "Sample rAte Converter". Is this the case or do you use something different to get better quality? AND...Do you convert the original raw audio track first prior to the bounce, or do you do it in another fashion?
  2. Dedric

    Dedric Guest

    In most cases with software sample rate conversion, any advantage of using 48k is lost when downsampling to 44.1k. Unless you are having a mastering house do the conversion, or are using a high end algorithm, I would suggest recording at 44.1k and avoiding the artifacts of SRC. This will also save you a step when burning to CD.

    If you decide to use 24-bit recording for live instruments or vocals, then Logic 5 has POW-r dithering for converting to 16-bit. This does work well, sounds very good, and doesn't have the same hazards as sample-rate conversion. There are noticeable benefits.

  3. pan

    pan Guest

    jscott wrote:
    Definitely stay on 48k, where you started and convert the 16bit/48k-Bounce with a dedicated SRC-program! I made good experience with Barbabatch?I think it's called...logic's SRC doesn't cut it...IMHO

    Dedric wrote:
    ...sorry Dedric, but you are combining things, you'd rather not mix up:

    Bit Depth, Dithering, Sample-Rate and SRC

    Dither is applied, when you change the Bit Depth, i.e. from 24 to 16 or in a Fadeout. 24 Bit give your recording a better DYNAMIC resolution (s/n-ratio and such) than 16 Bit, while 48k Sampling-Rate gives you a better FREQUENCY resolution than 44.1k.

  4. Dedric

    Dedric Guest

    Sorry I wasn't more clear Niko. I do know the difference very very well. I was attempting to point out by comparison, that there are advantages of 24-bit over 16 bit, and the conversion isn't nearly as damaging as SRC is for downsampling. The frequency response advantage of 48k is minimal (8.8% increase in frequency response to be exact), where the dangers of using a poor SRC program or algorithm are much more audible.

    Bit depth yields more range for perceived depth and clarity in a track or mix - esp. of decaying frequencies (reverbs, ambience). Even higher frequencies at lower levels become more noticeable, adding to the "clarity" perception even if the sampling frequency wouldn't seem to imply that gain.

    Sample rate determines captured frequency response (which also translates to clarity, but for different reasons). Since the Nyquist frequency of 44.1k is above human hearing, we are talking about harmonics and how the upper frequencies are handled when processed in a mix.

    I've also tested several SRC algorithms and I haven't found many that don't damage the high end (artifacts). Logic's SRC isn't good. Prosoniq's SonicWorks SRC seems pretty good, but still not sufficient for converting a mix compared to starting at 44.1 and staying there. I haven't tried Barbabatch, but will. I have heard good things about Sequoia and Samplitude uses the same audio engine (I'll be checking soon to see if the SRC is the same for 96k work).

    I, and many other engineers and producers I know believe (from hearing) that 48k isn't enough of a gain to risk SRC artifacts, where increasing the bit depth gains some of that perceived "clarity" back in dynamic range (particularly ambience in acoustic instruments) by using 24-bits and dithering the final master to 16. In the end I think 192k and 24-bits will be the best format, but the processing and storage tradeoffs are very large, esp. for larger multitrack projects. Eventually though I am sure our DAWs, cpu's and disks will support it with no problem (as many currently do), and our ears will appreciate it.

    Well, there's my theory on the subject. And you didn't even ask for it.

    Here's to dithered bits of terminology and sample-rate converted posts,
  5. pan

    pan Guest

    ...keep them coming! I prefer dithered posts, btw - no artifacts! :D

  6. pan

    pan Guest

    P.S: I like the UV22 better, it doesn't add noise...or just less, or more other pleasant artifacts.

    But we got away from the SRC-question Scott/(Jill?) :roll: asked. For some - especially film/video projects, it is suitable to start with the SR, that will be used for the mix. Here in europe it's usually 48k i.e. AVID/Protools.
    - But my soundlibrary is 44.1, so I do work in 44.1 and convert the bounces in the end to 48k, perhaps play them to a timecode-DAT at 48k. This is the problem the other way 'round, but the dithered theory would be, to

    ¡(great¿)Stay with your original material as long as possible, before doing degrading SRC!

    I have another question about bit conversion, and when to apply it:

    Is it better to stay with your original 16Bit material, or better convert it to 24Bit before doing things like EQ and fancy plugins? What's your opinion, gals?¿

  7. Dedric

    Dedric Guest

    Have you compared it to Logic 5's POW-r? So far POW-r (with noise shaping) seems very good, but I haven't put it to any extensive torture tests.

    I look at it as two separate aspects to bit depth requirements:

    First you have the original signal and the role bit depth plays in capturing the nuances of that signal - i.e. 24-bit for acoustic guitar can capture more ambience than 16, but for an SM57 on a Marshall amp with electric 24 is probably a waste of bits.

    Second, for processing in a mix, the bit depth at which the actual DSP runs affects the headroom (esp. with EQ and reverbs). At that point, whether the audio file is 16 or 24 bit, I think it is better to have 32-bit float or at least 48-bit fixed to allow for summing. I don't think you buy anything by converting 16-bit files to 24-bit before integrating into a mix, other than using more disk space. :D Usually in a mix you aren't actually recreating that audio file with any process that is saved backed to disk, but rather summing it with others. In this situation you would only pad it with 0's by converting to 24 bit - no new information. The processing of the mixer itself would convert it to 32-bit float (host-based) or 24-bit/48-bit fixed (ProTools). If you render or bounce an audio file back to disk with EQ or other DSP, then rendering to 24-bit can be of value as you would retain more of the processing depth of the DSP, with lower levels of dithering artifacts. I don't like adding dithered files together (as happens when bouncing submixes and track rendering), but sometimes it is unavoidable.

    Adding to the tangent of questions - anyone else want non-realtime track rendering in Logic? Paris has it - rendering a track with plugins takes about 5 to 10 seconds for a full 5 min track - really sweet. It would be especially nice to have in Logic. Maybe in 5.x. There is a freeware plugin that does this (in realtime I think), but it seems to have some bugs in Logic 5 on my system.

  8. jscott

    jscott Guest

    I asked the question on Sonik's site and the response came back as follows:

    I used the first method of doing the mix with Bounce+add and then converted the mixed track. The result was as predicted, slightly detuned and a bit round in tone. I have been told now on several different forums that it would appear, that as the popularity of ever high sample rates emerges, that going outside via analog to the likes of a Masterlink will yeild superior sound. Further, the idea scenario is to use the likes of an outboard clock, like Lucid or Ardvarkk?

    Would you all agree with that?
  9. Dedric

    Dedric Guest

    One mastering engineer I know prefers to choose the conversion technique based on the material. His preference is for SRC within Sequoia or Sonic Solutions depending on the material and sample rates (usually 96k or 192k to 44.1), but also goes outboard to high end converters at times.

    I have heard good things about Masterlink's converters. Some people seem to like the idea of going out D/A to A/D and others prefer to stay all digital with a really good SRC. It may come down to preference - there are tradeoffs with either method, technically speaking at least.

    Check user ratings on the converters or SRC you would be using, and then trust your ears. That's my approach anyway, FWIW. If you are on a Mac, give Barbabatch a try (per Niko's recommendation). There is a demo for it. If you are on PC, Samplitude Master 6.0 seems to have a pretty good SRC (although in the demo version the highest quality conversions don't seem to work on my system, so I don't know how it stacks up). Supposedly it has the same audio engine as Sequoia, although I would be surprised if the SRCs are the same considering that SM is about $230 and Sequoia is about $2500.

  10. pan

    pan Guest

    If possible, I'd go D/A A/D (That's what most Mastering Houses do anyways ;-)as mentioned above - if you can afford a Masterlink, go for it! And having a dedicated Masterclock for your DAW is really essential!


Share This Page