Skip to main content

sum of stems != original in Traktor Pro 4.1.1 (at least)

· One min read

Picture Did a small experiment in Traktor Pro 4.1.1

  • I created a .stem.m4a file that has "silent" drums, even though the original has drums.
  • Loaded into deck as standalone .stem.m4a file and tried messing with individual stem controls (volume, filter) - the track DOESN'T play DRUMS. This confirms that "sum of stems != original" when played as standalone .stem.m4a
  • Generated native TP4 STEMS, and then replaced generated .stem.mp4 file with my file (including necessary file metadata to play in TP4). Result is the same: the track DOESN'T play DRUMS.

TLDR: "sum of stems != original" in Traktor Pro 4.1.1 (at least), so if you truly want to play "complete" song - you need to generate stems using NUO-STEMS 🤷‍♂️

p.s. I didn't install 4.2.0 because it has some bugs related to files relocation that cause crashing, afaik

Release 3.1.0-beta.6

· One min read

3.1.0-beta.6 - 8 December 2024

Updates:

  • Fixed "Traktor Utilities" issue when loading collection showed no tracks (caused by Beatport tracks)
  • Fixed issue adding hidden files to the Queue. Happened when users selected folder to process & it contained hidden metadata files that don't contain any musical information (e.g. ".track.mp3"). Now all hidden files (starting with "." are omitted when adding to the Queue)

Release 3.1.0-beta.5

· One min read

3.1.0-beta.5 - 30 November 2024

Updates:

  • Fixed issue when second track in the Queue was always throwing error

Release 3.1.0-beta.4

· One min read

3.1.0-beta.4 - 29 November 2024

Updates:

  • Fixed issue when selecting CPU & number of cores wasn't working properly, leading to always using 50% of all available cores.
  • Updated Intel-based Mac implementation. Potentially leading to more AMD GPUs being supported (to be tested by users)

Release 3.1.0-beta.3

· One min read

3.1.0-beta.3 - 19 October 2024

Updates:

  • Fixed issues with adding FLAC & AIFF tracks to the Queue

Release 3.1.0-alpha.33+32

· One min read

3.1.0-alpha.33+32 - 11 September 2024

Updates:

  • New AAC Encoder on Windows -- QAAC (instead of FFMPEG aac). On Windows, NUO-STEMS is now using QAAC as AAC encoder. In simple terms: 256 AAC compression of stems shall be significantly improved resulting in higher quality. QAAC provides results that are almost identical to Mac AAC encoding.
  • NEW Feature: "AI distribute on 'drums' and 'other' stem" -- when using "sum of stems = original", new option is now selected by default.
    • It substantially cleans up "other" stem from leftovers of drums & percussion, while keeping "sum of stems = original"

Fixes:

  • fixed issue of not being able to convert using htdemucs & htdemucs_ft algorithms (resulting is "-1%")

Release 3.1.0-alpha.31

· One min read

3.1.0-alpha.31 - 4 September 2024

Updates:

  • Fixed a bug where conversion got stuck on "-1 %" if lyrics of the track contained special symbols.

  • 3.1.0-alpha.31 -- VERY IMPORTANT UPDATE.

    TLDR: You need to re-convert your STEMS if you used mel_band_roformer before.

    So, I discovered a bug that quality was not improved prior to this release, and it stayed the same as 3.0.0 for drums, bass & other.

    What it means (I'm comparing PRE-3.1.0-alpha.30 and 3.1.0-alpha.31):

    SDR (quality metric - check this video for explanation: (https://www.youtube.com/watch?v=NxphQmFIlk0&pp=ygUJbnVvIHN0ZW1z) ):

    • Drums: improved: 11.26 > 11.77
    • Bass: improved: 12.04 > 12.38 (you can actually hear that there are now very little drums leaks)
    • Other: improved: 5.74 > 6.90 (also, drums leaks are now minimal. Another bonus: of you enable "sum of stems = original", it now goes into "other" stem quite well)
    • Vocals: 10.85. (nothing changed here since release of mel_band_roformer)

    What it all means, is that quality improved substantially in different areas, and if you used "mel_band_roformer" algorithm to convert to STEMS since it's release, you now need to re-convert all your tracks to actually gain quality increases in all stems, and not just vocal