title: Does Audio Format Affect Stem Separation Quality?
subtitle: SDR benchmark across 6 audio formats on 20 MUSDB18-7s test tracks using
  htdemucs_ft
run_date: March 14, 2026
run_timestamp: '2026-03-14T11:47:24'
model: htdemucs_ft
dataset: MUSDB18-7s
dataset_url: https://zenodo.org/record/1117372
n_tracks: 20
stems:
- vocals
- drums
- bass
- other
key_finding: MP3 128kbps input reduces mean SDR by 0.24 dB compared to WAV 24-bit
  (7.8 dB vs 8.04 dB)
formats:
- id: wav_24
  label: WAV 24-bit
  category: lossless
  sdr_vocals: 8.86
  sdr_drums: 9.95
  sdr_bass: 9.52
  sdr_other: 3.85
  sdr_mean: 8.04
  delta_from_wav24: 0.0
- id: wav_16
  label: WAV 16-bit
  category: lossless
  sdr_vocals: 8.86
  sdr_drums: 10.01
  sdr_bass: 9.48
  sdr_other: 3.82
  sdr_mean: 8.04
  delta_from_wav24: 0.0
- id: mp3_320
  label: MP3 320kbps
  category: lossy
  sdr_vocals: 8.82
  sdr_drums: 9.93
  sdr_bass: 9.52
  sdr_other: 3.69
  sdr_mean: 7.99
  delta_from_wav24: -0.05
- id: mp3_192
  label: MP3 192kbps
  category: lossy
  sdr_vocals: 8.78
  sdr_drums: 9.92
  sdr_bass: 9.42
  sdr_other: 3.82
  sdr_mean: 7.98
  delta_from_wav24: -0.06
- id: mp3_128
  label: MP3 128kbps
  category: lossy
  sdr_vocals: 8.49
  sdr_drums: 9.69
  sdr_bass: 9.46
  sdr_other: 3.57
  sdr_mean: 7.8
  delta_from_wav24: -0.24
- id: aac_256
  label: AAC 256kbps
  category: lossy
  sdr_vocals: 8.78
  sdr_drums: 9.91
  sdr_bass: 9.57
  sdr_other: 3.69
  sdr_mean: 7.99
  delta_from_wav24: -0.05