illustrate
Products            Buy            Support Forum            Registrations            Professional            About           
 

Conversion issue with 32 bit float and volume normalize

Collapse
X
 
  • Time
  • Show
Clear All
new posts
  • bigjohn

    • Jul 2015
    • 9

    #1

    Conversion issue with 32 bit float and volume normalize

    I have been using dBPoweramp R14 music converter for many years. I recently replaced my PC with a new laptop and decided it was time to update dBP to the latest version. So I now have dBP 2025-07-01 64 bit on Windows 11. The first time I have tried to use this to process some music files I am getting badly distorted (unlistenable) output files. They are playable but consist of loud noise. Looking at the waveform in Audacity they contain square waves between 0 and -1 (i.e. no waveform positive above zero) - see screenshot below. The file sizes are also not correct (much smaller) based on comparing to previous output files that are ok.

    Click image for larger version

Name:	image.png
Views:	17
Size:	29.2 KB
ID:	335552

    The operation I am doing is to convert 6 channel FLAC files to 2 channel FLAC files using these DSP Effects:
    Bit Depth: -depth="32 float"
    Channel Mapper: -channels="2" -ch1="1,-3,3,-6,4,-3,5,-3" -ch2="2,-3,3,-6,4,-3,6,-3"
    Volume Normalize (non-live): -mode="fixed" -maxamp="8" -desiredb="0" -adapt_wnd="6000" -fixed="-5" -eburefdb="-18"
    Bit Depth: -depth="24"

    I have done that many times in the past using dBP R14 without any issues.

    From playing around with this to try to find the cause, I don't think the Channel Mapper is significant, it seems to be an interaction between the 32 bit float Bit Depth and the Volume Normalize.

    If I do this, the output file is OK:
    1. Bit Depth - 32 bit float
    2. Bit Depth - 24 bit fixed

    If I then add a volume normalize step in the middle of that, I get distorted incorrect output file:
    1. Bit Depth - 32 bit float
    2. Volume Normalize - Fixed Amplification -5dB
    3. Bit Depth - 24 bit fixed

    If I then change the first step from 32 bit float to 32 bit fixed and leave steps 2 and 3 unchanged, the output file is OK:
    1. Bit Depth - 32 bit fixed
    2. Volume Normalize - Fixed Amplification -5dB
    3. Bit Depth - 24 bit fixed

    Please help!

    I have captured the debug log file from one of the runs where it produces the distorted output file. I don't see any error messages or anything that looks wrong in the log file.
    Attached Files
  • Spoon
    Administrator
    • Apr 2002
    • 45401

    #2
    We are working heavily in this area (new ARDFTSRC resampler), so will test this. It looks like volume normalize is having issue with floating point, which is strange.

    I am not getting this issue though.

    As a test, take a 2 channel CD quality FLAC file (16 bit, 44KHz):

    do your conversion to wave, with:

    Bit depth - 32 bit float
    volume normalize fixed -5b
    bit depth - 24 bit (no dither)

    does that work? There must be something else causing it.
    Spoon
    www.dbpoweramp.com

    Comment

    • bigjohn

      • Jul 2015
      • 9

      #3
      I agree that there must be some kind of weird interaction going on to cause this.

      So, testing it with a 16/44.1 2-channel FLAC file as input gives a valid output file, same for a 24/48 and a 24/96 2-channel FLAC files as input. That is doing these steps as suggested:
      Bit depth - 32 bit float
      volume normalize fixed -5b
      bit depth - 24 bit (no dither)

      The 6 channel FLAC file (which is 24/96) that I have been using is one that I have previously processed successfully using dBP R14 on my old PC. Using my new laptop I can successfully open and play that 6 channel file in both Audacity and VLC, so I don't think the file itself is inherently faulty. Plus I have tried processing some other 6 channel FLAC files in dBP with the same result. Also, using the 6 channel FLAC file as the input and trying different types of Volume Normalize produces different results:
      Volume Normalize - Fixed Amplification -5dB -> stereo output file is distorted, size 2MB
      Volume Normalize - Peak to Peak desired volume -1dB -> stereo output file is distorted, size 6.5MB
      Volume Normalize - Adaptive [Peak to Peak] desired volume -1dB -> stereo output file is ok, size 26MB

      Does that suggest anything about what might be the cause of this?

      Those are back to the original process, i.e.:
      Bit depth - 32 bit float
      Channel mapper 6 channel to 2 channel
      volume normalize
      bit depth - 24 bit (no dither)

      Comment

      • Spoon
        Administrator
        • Apr 2002
        • 45401

        #4
        Are you able to email your file? (flac file before conversion, if your email cannot send, upload to dropbox or similar and provide a link please):


        Spoon
        www.dbpoweramp.com

        Comment

        • bigjohn

          • Jul 2015
          • 9

          #5
          I've sent an email with a link to the file

          Comment

          • Spoon
            Administrator
            • Apr 2002
            • 45401

            #6
            We can reproduce the issue and will fix for next release. It is the 6 channel >> 2 channel which is causing the issue.
            Spoon
            www.dbpoweramp.com

            Comment

            • bigjohn

              • Jul 2015
              • 9

              #7
              OK, thank you, I'll wait for that release.

              Comment

              Working...