Music Production

Both denoisers were run over a clean programme in which there was nothing to remove, so that everything missing at the output is damage. The noise reduction control barely changes anything, while the noise floor control moves the result by 20.9 dB, and the best setting lands within 1.6 dB of the real noise floor every time. Below a noise level of 40 dB under the signal, the best achievable improvement is 0.15 dB.

The mix sounds good, and there is one more decision before export: a clipper, a limiter, or a combination? Instead of relying only on the louder impression, it helps to compare at the same loudness – including the…

All 23 crossfade curves of ffmpeg measured: seven hold the level when both sides carry the same take, two hold it when they carry different recordings, and none holds both, so the wrong choice costs 3.01 dB either way. A cut exactly on a zero crossing still clicks at −37.44 dBFS when the slope reverses there, and the fade needed to bury the click does not depend on the frequency of the material.

Three ways of widening a mix, measured with ffmpeg on a 32 second piece: raising the side costs nothing in mono beyond its own level, 0.30 to 1.10 dB spread evenly across the spectrum, but does nothing at all to a mono source. Delay based widening reaches a correlation of 0.04 or tears eight twelfth-octave bands more than 6 dB out of the mono sum, the deepest 9.37 dB at 50 Hz.

A master delivered at −1 dBTP rarely reaches listeners in that form: encoding to AAC, MP3, Ogg Vorbis or Opus can raise its peaks. Across 88 encodings with ffmpeg, heavily limited masters at −9 LUFS and −1 dBTP lost up to 0.98 dB of headroom, and one AAC file went above 0 dBTP at a single spot. More dynamic masters at −14 LUFS stayed at or below −0.84 dBTP, a loud master limited to −2 dBTP at or below −0.89 dBTP. With the codecs the streaming services use and the recommendations of Spotify, AES and EBU.