bouldersound, post: 449370, member: 38959 wrote: All done at 44.1kHz/16 bit.

Many modern converters have headroom in the analog parts of the circuit to accommodate intersample peaks, though I think that's technically a violation of the Redbook standard.

That's interesting about the converters and intersample peaks.

bouldersound, post: 449370, member: 38959 wrote: Many modern converters have headroom in the analog parts of the circuit to accommodate intersample peaks, though I think that's technically a violation of the Redbook standard.

Many people use Prism converters for that reason.

audiokid, post: 449372, member: 1 wrote: Many people use Prism converters for that reason.

That's nice for the person listening through the Prism converter. What happens if they are producing audio that distorts on other converters but can't hear it on their system? This is one exception to the "use your ears" guideline. One tool is this: http://www2.solidstatelogic.com/news/ssl-unveil-x-ism-innovative-inter-sample-meter-plug-and-make-it-available-free-download

Of course, using the best converters you can is going to be of benefit, there's no downside to that.
But if people are relying on a service like SoundCloud to do the conversion for them, or, if SC is re-converting the uploaded files, then it's touch and go as to whether that conversion is done well.
I'm one of the members here on RO who has heard obvious lossy artifacts (and have also seen peaking) on files after an upload to SC has been done. (I'm pretty sure Marco has experienced this with SC as well (pcrecord )
Around a year ago, when I finally decided to ditch SC for good and use ROs media player exclusively, the event that made that happen, was a file I'd uploaded to SC, and then imported back into Samp... the file had 5 separate and distinct "overs". While the lossy artifacts I had heard on other SC uploads were still audible (the phasy/swirly thing happening on the top end ) the peaks didn't cause any noticeable distortion, maybe because they happened so fast, but ...the Overs were still there, and I was able to see them on both Samp's meter and on another separate 3rd party meter as well. Boulder (bouldersound) mentioned the potential for distortion to be heard on other systems. For me, even if there's no apparent or obvious audible distortion, the clipping still bothers me, because those overs aren't supposed to be there, as my final MP3 render on the original mix on Samp had a LUFS of -16db with an absolute peak level set at -.5, using the S-Max limiter that is stock in Samplitude.
We need to remember that no matter how much care we take, or how good our conversion is, the file will still be at the mercy of wherever (or whomever) we send it to (speaking particularly of online distribution and streaming).
Since that event, when the destination of a mix is going to involve uploading an MP to any online service, I've allowed an extra -1 DB of headroom on the final mix when outputting to MP3, regardless of the resolution of the MP file.
And... I don't use SoundCloud anymore because I just don't trust them to not drastically alter the file, and ultimate the quality.
IMHO of course.

DonnyThompson, post: 449375, member: 46114 wrote: I'm one of the members here on RO who has heard obvious lossy artifacts (and have also seen peaking) on files after an upload to SC has been done. (I'm pretty sure Marco has experienced this with SC as well (pcrecord )

I didn't experience it myself. I still have a few files on SC and they sound ok.. But I hear the phasyhighs half of the time on RO member posts. Wouldn't it be fun to now prove that they happen because the original file was too loud for the SC convertion ??

When I get time, I'll test that !

DonnyThompson, post: 449375, member: 46114 wrote: While the lossy artifacts I had heard on other SC uploads were still audible (the phasy/swirly thing happening on the top end ) the peaks didn't cause any noticeable distortion, maybe because they happened so fast, but ...the Overs were still there,

Doesn't samplitude have a special way of dealing with the levels and overs ?? I think I read something about that.

You have to be careful when talking about the effects of input peaks with output peaks in correctly-configured equipment.

The Shannon sampling theory says that in a signal that is band-limited (i.e. in frequency) to less than half the rate at which it is sampled, a sampling process can always accurately represent the content in both amplitude and frequency. One often-misunderstood result coming from this is that it is necessary for the analogue input section of the ADC not to saturate, i.e. that analogue saturation levels are greater than the d.c. level that causes full-scale digital values. It's the argument about how much constitutes "greater than" in this context that can differentiate between good converters and excellent converters. Without plodding through the theory again, my memory is that the ratio is not huge, and may be about 1.1 times. The point is that the anti-aliaising filters that are used to obey the Shannon restriction inhibit the possibility of values going any higher than this figure between samples without showing overload at the ADC. Things are a little more complicated when over-sampling ADCs and DACs are used, but the principle still applies.

In one of my commissioned designs for a certain commercial product, I was asked specifically to research this over-range subject and to make sure at both the design level and in prototype test that the ADC values would represent the full range of the analogue input, whatever pathological waveform was presented to it. I duly did some investigation, but wasn't told why it was necessary to make a big fuss over it. Much later, I saw the way the marketing department had turned it into a unique feature, which of course it was not, as any competent design would perform in this way.

When it comes to D-A conversion, there is a similar but different problem. If digital values arrive at the DAC representing an analogue waveform that has momentary levels value above the d.c. FS level, how should the DAC behave? Again, it's in the filtering. The DACs can only convert values within their digital FS range, but when a sequence of values is put through the reconstruction filters, the resulting values at the analogue filter output can have peaks that exceed the d.c. digital full range. This means that the designer has to ensure that both the reconstruction filters and the output amplifiers operate with enough headroom to pass these peak values to the output socket.

It's my understanding, conceding that I'm not terribly certain about it, that when digital audio started intersample peaks simply could not happen. There was no audio DSP, only direct ADC to DAC. CD masters would be made from analog sources by engineers who knew that they simply could not exceed the FS limit as that would result in either analog clipping or nonsense digital data that would be converted to harsh noise. Under those conditions it was impossible to have a series of samples that, when reconstructed, could drive the DAC to voltages past its limit. Then along came DSP audio production and the potential was opened for data that represented voltages beyond the DAC's limit. Since then many converters have been designed to accommodate ISP, though technically they violate the Redbook standard, which requires that V-out equals V-in.

I don't think the advent of affordable DSP has made much difference. Even in the early days of CDs it was possible to have replay waveforms whose peaks exceeded the nominal output of the D-A converter chips because the detail of the shape of the output waveform is generated by the reconstruction filter, not directly by the DAC. I remember there being articles in hi-fi magazines of the day showing the analogue outputs of even some high-end converter boxes clipping because the designers had failed to allow for this effect.

Why wouldn't the ME provide you with an MP3, a 16 bit, and a 24 bit file? You shouldn't have to convert anything, he should have provided you with those formats when he handed over the final product. I mean granted I'm not going to do an SACD conversion, but at this stage of the game, certain formats should be expected from an ME. That includes 24 Bit Wav, 16 Bit, and 320 MP3. On occasions I'll get asked for an AIFF file. Part of the ME doing his job is giving you standard formats to avoid any sort of errors or having to convert after the fact. Strange...

Not to beat a dead horse here but

The FORMAT can - as Keith (Keith Johnson ) mentioned - intersample peaking has been known to occur in the conversion from wave to MP3 ... but that's caused more by the codec "rounding" down or up; so Keith mentioning that a headroom of 2db -3db being allowed for that format is a good idea...

If the ME did his job and send you all the file formats printed from the source material, why would you need to convert anything?