Sound Quality in Release 4.2

[Skip Pack]

I've had 4.2 up for about a week now, and until today, felt it sounded like 4.1, good but not as nice as 3.8.4 LL. I had upsampling set to 384/24.  I have an Allo Boss Dac 1.0, and am sending max resolution data to it. It has always sounded better, avoiding the on-DAC-chip upsampling.  Earlier today I switched the  SoX upsampling to 384/32, and it sounded a good deal better. Theoretically the 32-bit words just have eight more 'zeros' added to the 24 bit  upsampled result, and I can't see how that could make such a clear difference. I went back and forth several time, and the difference in obvious.

My only guess is that there may be oversampling induced clipping going on when it upsamples to 384/24.  From earlier discussions I have gotten the impression from Tim that the SoX parameters in the calls that MPD makes are not fully available the the programmer making the MPD calls. I experienced this effect once using SoX from the command line. The upsampling command offers an option to lower the whole signal a specified number of db in order to allow room for the peak samples that are interpolated at a higher level than the input stream (something that has to happen increasing the sample rate at peaks). The following is a call to the SoX upsampler:

sox $INFILE -b 24 $OUTFILE gain -n -3 rate -m -p 20 384k 

The -n option here is setting the output stream 3 db lower. I had tried this line without the -3 setting and it wasn't good, so I fiddled to the '-n -3' setting which seems fine. 

The good news, for me, is that I'm now happy to leave 3.8.4 behind as a fond memory and enjoy the many improvements in MoOde and use the RPi3+.

I have no idea about any effect of this consideration on DACs that don't turn off on-board resampling when receiving 383/x streams.

Skip

[rhizomusicosmos]

I've had 4.2 up for about a week now, and until today, felt it sounded like 4.1, good but not as nice as 3.8.4 LL. I had upsampling set to 384/24.  I have an Allo Boss Dac 1.0, and am sending max resolution data to it. It has always sounded better, avoiding the on-DAC-chip upsampling.  Earlier today I switched the  SoX upsampling to 384/32, and it sounded a good deal better. Theoretically the 32-bit words just have eight more 'zeros' added to the 24 bit  upsampled result, and I can't see how that could make such a clear difference. I went back and forth several time, and the difference in obvious.

My only guess is that there may be oversampling induced clipping going on when it upsamples to 384/24.  From earlier discussions I have gotten the impression from Tim that the SoX parameters in the calls that MPD makes are not fully available the the programmer making the MPD calls. I experienced this effect once using SoX from the command line. The upsampling command offers an option to lower the whole signal a specified number of db in order to allow room for the peak samples that are interpolated at a higher level than the input stream (something that has to happen increasing the sample rate at peaks). The following is a call to the SoX upsampler:

sox $INFILE -b 24 $OUTFILE gain -n -3 rate -m -p 20 384k 

The -n option here is setting the output stream 3 db lower. I had tried this line without the -3 setting and it wasn't good, so I fiddled to the '-n -3' setting which seems fine. 

The good news, for me, is that I'm now happy to leave 3.8.4 behind as a fond memory and enjoy the many improvements in MoOde and use the RPi3+.

I have no idea about any effect of this consideration on DACs that don't turn off on-board resampling when receiving 383/x streams.

Skip

Your upsampling impressions are interesting as I identified there were clear unsuppressed alias spectra in the output of another PCM5122 IIS DAC, the HifiBerry DAC+ Pro (see http://moodeaudio.org/forum/showthread.p...149&page=4). Upsampling with SOX is likely to avoid this.

[Tim Curtis]

I've had 4.2 up for about a week now, and until today, felt it sounded like 4.1, good but not as nice as 3.8.4 LL. I had upsampling set to 384/24.  I have an Allo Boss Dac 1.0, and am sending max resolution data to it. It has always sounded better, avoiding the on-DAC-chip upsampling.  Earlier today I switched the  SoX upsampling to 384/32, and it sounded a good deal better. Theoretically the 32-bit words just have eight more 'zeros' added to the 24 bit  upsampled result, and I can't see how that could make such a clear difference. I went back and forth several time, and the difference in obvious.

My only guess is that there may be oversampling induced clipping going on when it upsamples to 384/24.  From earlier discussions I have gotten the impression from Tim that the SoX parameters in the calls that MPD makes are not fully available the the programmer making the MPD calls. I experienced this effect once using SoX from the command line. The upsampling command offers an option to lower the whole signal a specified number of db in order to allow room for the peak samples that are interpolated at a higher level than the input stream (something that has to happen increasing the sample rate at peaks). The following is a call to the SoX upsampler:

sox $INFILE -b 24 $OUTFILE gain -n -3 rate -m -p 20 384k 

The -n option here is setting the output stream 3 db lower. I had tried this line without the -3 setting and it wasn't good, so I fiddled to the '-n -3' setting which seems fine. 

The good news, for me, is that I'm now happy to leave 3.8.4 behind as a fond memory and enjoy the many improvements in MoOde and use the RPi3+.

I have no idea about any effect of this consideration on DACs that don't turn off on-board resampling when receiving 383/x streams.

Skip

Hi Skip,

Thats correct. MPD only exposes a basic set of SoX capabilities (recipes) including the 3 quality settings, sample rate and bit depth, and a system param to run SoX multithreaded. There are many more SoX recipes available via sox cmd line.

-Tim