Announcement

Collapse
No announcement yet.

Audio quality on SoundCloud

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

  • Audio quality on SoundCloud

    Hello -

    I just uploaded a track to SC, then played it back from the SC site - the sound quality was ... erm ... appalling :frown:. At least 60% of the audio information was missing and the rest was simply degraded. A result of converting to SC's default MP3 bit rate I guess.

    Does anyone have any tips for improving the sound quality of an uploaded track on SC?

    The only thing I can think of is that as the track is 70 minutes I rendered it as an MP3 320 and uploaded that. The 16bit WAV file is 790 mb. So my original file to SC was already compressed. I'm thinking that maybe when SC converts from WAV to MP3 the result is better than converting from MP3 320 to MP3 120.

    Any thoughts in this?

    A.

  • #2
    I am hoping some with more experience/knowledge chime in here, but to get the conversation started: I have done some of these things in the past and it seemed to help.

    I don't like getting deep into the math of things, so some of this may just be perceptual bias. Someone please jump in and correct any misinformation.


    • Try allowing more headroom and reducing the dynamic range of the mix going to SC. Avoid heavy limiting and compression. Mix a bit quieter for SC.
    • Convert your files to 128 bit mp3--see if you can detect any problems before going through uploading. I would still submit the lossless format file to SC though (wav/flac/aif, etc.)
    • Most of the problems I have noticed in my uploads were in the far ends of high and (most often) low frequencies. Try a low and/or high pass filter for your SC mix.
    • Distortion and saturation seem to attract problems in conversion. My mixes are intentionally dense and that usually causes trouble as well. I usually make sure links to my bandcamp versions of the song are included. (not a fix, but a workaround).
    • Avoid stereo widening processing for SC tracks.


    Hopefully someone can step in with corrections or explanations.

    :biggrin:

    Comment


    • #3
      As I understand it, each time, you transcode an audio file, some data will be lost. In your case, you have converted from WAV to mp3 (at 320K). This will cause some of the audio data to be lost. SoundCloud's encoder will transcode your 320K file to 128K, causing another loss of data. Uploading the WAV file to SoundCloud will mean that only one trancoding operation will be required, which, in theory will create a "better" mp3 than would be possible using 2 serialised transcoding operations. You should be able to verify this by performing the mp3 transcoding yourself and listening to the resulting audio. If you are going to upload 70-minute WAV files to SoundCloud, you will need a decent broadband connection to do this. Anything slower than 2.5Mbit/sec upload speed will be quite tedious.
      Whatsisname's Little Fluffy Clouds | Campsite | Hearthis | Orfium | SeismicTC | Twitter | Ello

      Comment


      • #4
        Thanks for the replies.

        I noticed that SC's files play at a pretty high volume level by default so, following one of aoVI's suggestions I did a new render of the file to FLAC with the peaks at -7db. Uploaded the FLAC file just now and checked it - the resulting sound quality is greatly improved. Two transcoding operations on the same file also results in significant loss of audio information. It seems that the two issues I've described were the cause of the problem.

        Comment


        • #5
          Out of interest, do you remember what the peak levels were on your original render? I'm looking at a problem on a heavy rock track for somebody else and the scenario you have described here may be similar. The track I am currently listening to may have problems on the hihat or crash cymbals.

          I think that it is possible that SoundCloud may perform some sort of level normalisation during its transcoding operations. Hearthis.at may be similar. I'm guessing here, but I think the input peak and/or RMS levels will have a major effect upon the output values. I don't think that the visual representation of the track waveform is particularly informative in this respect.
          Whatsisname's Little Fluffy Clouds | Campsite | Hearthis | Orfium | SeismicTC | Twitter | Ello

          Comment


          • #6
            Originally posted by aoVI View Post


            • Try allowing more headroom and reducing the dynamic range of the mix going to SC. Avoid heavy limiting and compression. Mix a bit quieter for SC.

            :biggrin:
            I think the key here is creating the headroom. Compressors and limiters will generally reduce DR within a track, but increasing the headroom via individual channel levelling should preserve DR as the decrease in individual channel levels should affect the min/max level ratio within each channel in a similar fashion.
            Whatsisname's Little Fluffy Clouds | Campsite | Hearthis | Orfium | SeismicTC | Twitter | Ello

            Comment


            • #7
              My earlier comment about headroom may not be correct. I've just been checking some of my recent tracks, and in almost all cases I have less than 2dB peak headroom with DR values between 10 and 15. My tracks are uploaded as 320k mp3 files, to allow a reasonable quality of download, so I am going through the same "double-transcode" exercise that was described earlier. The resulting 128k streams don't sound bad.
              Whatsisname's Little Fluffy Clouds | Campsite | Hearthis | Orfium | SeismicTC | Twitter | Ello

              Comment


              • #8
                it is sort of conflicting advice though. I am really bad at discussions involving compression, equalization, and dynamic range. I probably should have worded it:

                Try allowing more headroom and avoid using heavy compression and limiting. Mix more quietly for SC tracks.

                I never owned a compressor or limiter back in the analog age so I never learned to use them properly. Getting access to them in the box resulted in some really smashed mixes early on, which sounded even worse on SC. :canthearyou:

                Comment


                • #9
                  Originally posted by seismic1 View Post
                  Out of interest, do you remember what the peak levels were on your original render? I'm looking at a problem on a heavy rock track for somebody else and the scenario you have described here may be similar. The track I am currently listening to may have problems on the hihat or crash cymbals.

                  I think that it is possible that SoundCloud may perform some sort of level normalisation during its transcoding operations. Hearthis.at may be similar. I'm guessing here, but I think the input peak and/or RMS levels will have a major effect upon the output values. I don't think that the visual representation of the track waveform is particularly informative in this respect.
                  I didn't use compressors or limiters, just some EQ on some of the processed samples. I rendered the master mix to WAV 24bit then added a little spatial enhancement with SHEPPi. That was rendered to WAV 24bit. The peak levels on the original master were at approximately -6db and I normalised peaks to 0db in Wavelab. That was the extent of the mastering on the track. The normalising process in Wavelab did not affect the sound quality in any way.

                  I then rendered the final 24bit Wavelab master to 16bit WAV for CD (with peaks at 0db) which gave really good sound quality when played back from a CD on my hifi system. That was the file I converted to MP3 320 which I uploaded to SC and the result was pretty bad. Dropping the peaks to -7db and uploading a FLAC of that to SC resulted in a quite noticeable improvement in sound quality. It's not the best of course, being MP3, but I can live with it.

                  I think that the problem was caused by a combination of peak levels at 0db, and two transcoding operations on the audio file. As I mentioned previously, I did notice that SC playback levels are quite loud. I compared my WAV 16bit file and SC's MP3 file at the same volume setting on the computer. The MP3 streamed from SC was significantly louder.

                  Hope this helps...

                  Comment


                  • #10
                    I wouldn't expect normalising to affect the sound quality when comparing the Wavelab output to the input. My concern is that the mp3 transcoding can be affected by higher levels contained within the input data. I used to process seismic data (a long time ago) in the frequency domain via FFT and some of the algorithms were particularly sensitive to small parameter changes, which could have a significant impact upon the quality of the resultant wavelet. Things have obviously moved on since then, but I find it hard to shake the belief that ears are more credible than algorithms.

                    In case anybody is interested, here is a brief summary of the mp3 encoding process, although it raises as many questions as it answers:-

                    http://arstechnica.com/features/2007...compression/1/
                    Whatsisname's Little Fluffy Clouds | Campsite | Hearthis | Orfium | SeismicTC | Twitter | Ello

                    Comment

                    Working...
                    X