The SRC Wrap Up: Part IV
As I’ve explained over the past three days, converting from one sample rate to another can be accomplished by passing through an analog stage, out of real time using clunky brute force methods and in real time using powerful DSP processors and fancy algorithms. There are advantages and disadvantages to each of these conversion schemes…we’ve looked at a few of them over the past few days. But what if you wanted a “no holds barred” way to get from one sample rate to another? What could an uncompromising audio engineer do?
In the not too distant past you could purchase a highly tweaked piece of hardware from Harmonia Mundi Acustica or dCS (data conversion systems) that could convert 48 to 44.1 kHz. These machines were very expensive and specialized. I have rented these converters on several occasions and found them very capable. But as I learned a long time ago while studying computer science and electronics, anything that can be built in hardware can be done in software. This is a profound statement. It means that anything you can do with a dedicated device can be done with a bunch of microprocessors and some really inspired programming.
Re enter Sonic Solutions. The engineers in San Francisco realized that the real time SRC algorithm that was built into the loading section of their workstation would be good enough for the vast majority of conversions done at mastering facilities. And it was…I used it 99% of the time. But Andy Moorer, the co-founder and guru of audio DSP at Sonic, included a “background” sample rate conversion tool in their software. It works OORT (out of real time) as a background task using the power of the Motorola 56000 DSP chips that populated their sound cards.
Those of us that will accept nothing but the very best when it comes to our own projects, were quite pleased with the options that the OORT SRC tools presented. This optional method allowed users to increase the complexity of the algorithm used to do the conversion. I recall that the real time version had a multiplier of around 5-10 (a quality factor in the mathematical calculation)…but if you chose to use the OORT version, you could crank that number up to 200! Of course, it took forever to crunch through all the needed calculations but it was worth it.
The current version of Sonic Studio, which is no longer a part of the original company, has a dedicated application called “Process” to handle sample rate conversion among other processes. There is no need for specialized processors to augment the CPU in personal computers these days. The CPUs are so powerful that they can handle the conversions all by themselves. They aren’t done in real time but they are done with amazing quality. Don’t waste you money on hardware when you can simply write some code that will run on virtually any machine. Very smart.
Process allows engineers to tweak the SRC in a variety of ways. You can select the steepness of the filter (gentlest, gentle, steep) and the type of dither that you want (triangular, rectangular, noise-shaped or none) and the algorithm type (Sonic HD or classic).
I have painstakingly processed one of my favorite tracks using ALL of these options and carefully reviewed them in my studio. I took a portion of the award-winning Laurence Juber “Mosaic” track, which was recorded at 96 kHz/24-bit PCM and converted it to CD standard 44.1 kHz/16-bit PCM. I have put them all up on the FTP site and invite you to download them and listen to them. There are differences…but you will need a very quiet environment, a very good system or a set of excellent headphones to detect them. There is no worst or best…just pick the sound that you like. But remember the source recording is pretty spectacular.
So getting from one sample rate to another can be done a lot of different ways…and they matter to the final sound that you hear.
Not to beat a dead horse, but this discussion of downsampling got me thinking about a specific aspect of Steven Stone’s article, “DSD and PCM – Horses for Courses” that was mentioned in a previous post. From what I could understand from the article, the main reason that he would choose DSD over PCM is the “lack of flexibility” for PCM, specifically regarding when the final production was intended for redbook audio standards. Based on what you’ve been saying about sample rate conversion, these tools seem pretty advanced. So I’m confused about why the stated scenario of a non-integer downsampling of PCM from 192/24 format to Redbook would have reduced fidelity compared to a DSD-recorded 176/24 downsampled to Redbook format. Were these decimation errors an issue in the past when tools/computers weren’t as advanced, or is this a significant issue in today’s production workflow?
Oh BTW, I did listen to the original and downsampled Mosaic tracks. I had listened to the high resolution version in the past and it sounded detailed as I recall. The CD quality version also sounds equally good, and personally with my hardware (Dragonfly DAC and Sennheiser HD558 headphones) I couldn’t pick up any differences. Though my computer computer case fans are a bit loud!
It used to be that you could make a reasonable case that integer multiple sample rate conversions did a better job than non-integer downsampling. However with the advent of newer algorithms, faster processors and higher quality DACs, even non-integer conversions are being done with extremely high quality. I wouldn’t quibble with Steven about his choice to record using DSD if he knew that the end product would be a compact disc as long as he’s aware that he won’t be able to do any post production processing on the recording until he got it converted to 44.1 kHz/16-bits. And if you know that you’re producing for a CD release, why would you record using DSD in the first place?