Audio Asylum Thread Printer

In Reply to: RE: Christie Winn - C'est Magnifique posted by Roseval on September 11, 2011 at 07:17:32

As converted by foobar2000 to 176.4/24 there is music related energy up to about 31 kHz. This can be seen in the screen shot of iZotope RX 2 Advanced. One can clearly see the musical energy in the two plots, one where intensity is encoded by color and the plot is time vs. frequency and the other a more traditional spectrograph at one part of the recording, where I have superimposed this separate window. (Click on the image if you want to see a higher resolution version.)

It looks like the foobar2000 DSDIFF module got the filtering approximately right. I note that the transition of the filter starts somewhat below 88.2 Khz, which prevents residual DSD noise from aliasing while still preserving audio signal. If one expects much more than 40 Khz bandwidth through DSD one is kidding oneself as the damn 1 bit noise creates problems. If left in it cruds things up and if filtered out it takes music with it. (In case you didn't guess, I am not a fan of DSD and DSD to PCM conversions are the worse of both formats as one loses S/N ratio thanks to DSD and one loses transient response and/or bandwidth thanks to the lower sampling rates of PCM.)

Unfortunately there is a problem with the foobar2000 converter. The gain used to convert from DSD to PCM is set a bit too high. While PCM has a hard limit on the maximum undistorted sample, this is not the case with DSD, where the engineer can go a few dB over "maximum" DSD level at the penalty of a small amount of distortion. As a result there is no "correct" setting for converter gain. However, with 24 bit PCM there is more dynamic range than one gets with DSD, so the conversion should be set conservatively. This appears not to be the case with the foobar2000 converter. Of the 12 tracks on "C'est Magnifique" the conversion clipped two of them. One track was clipped for about 10 samples (at 176.4 kHz) on a piano note and this was not audible. Another track had two clips that went on for 18 samples and there was definite vocal harshness which was audibly recognizable and which I had noted while listening to the music before looking at the actual samples with Soundforge. The audible clipping occurs on track 6 at approximately 3:22. From using the "declip" functionality of iZotope RX 2 Advanced it appears that this clipping could have been avoided if foobar2000 had used 0.52 dB of attenuation over and above whatever it has built-in.

I must emphasize that this is a problem with foobar2000, not with the recording. IMO the gain in foobar2000 for converting DSD to PCM should be taken down another 1 or 2 dB. Whenever a sample rate conversion is performed there is the possibility that peak PCM levels may be exceeded. In the case of PCM input this can easily be dealt with by the user by attenuating the input signal, but there is no way to attenuate DSD without converting it to PCM. Consequently to avoid possible distortion DSD to PCM converters should either provide very conservative built-in gain or come with a user-adjustable gain settings. (Weiss Saracon takes the later approach.) In the past some commercial downloads produced from SACDs have had clipping problems and these were eliminated by corrections to the conversion software.

Most audiophiles will probably not notice this clipping, by the way. However, any competent recording engineer will notice it immediately unless they have ruined their hearing by producing over-compressed music by prostituting themselves to the dollar in the loudness wars. The ability to hear common forms of distortion is an essential (learned) skill for a recording engineer as this can happen from misuse of tools or equipment defects.

Tony Lauck

"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar

Computer Audio Asylum

More on converting DSD to PCM

Post a Message!