top of page

The Master Guide / 13 Sound design and mix · The mix

Checking the mix on small speakers

Before you approve a mix, measure whether the words survive on small speakers and in mono. The figures are flags that tell you where to listen; none of them passes or fails a mix.

Before you approve a mix, measure whether the words survive on small speakers and in mono. The figures are flags that tell you where to listen; none of them passes or fails a mix.

The checks are [proven] as mixing craft; their numbers are [best available, not yet validated]. Run them on every review render and film master, beside the loudness line. An LU (loudness unit) is a step of 1 dB on a loudness meter.

1. Speech against the background, during speech

Export the stems (see Sound design and mix → Stems) and measure each stem's integrated loudness over the same stretch of speech, then subtract the background's from the speech's (steps below). Do it for each stretch of speech, and note the lowest margin across the stretches beside the flag line. Flag:

  • speech less than 10 LU above the music;

  • speech less than 15 LU above the ambience;

  • a voice less than 20 LU above another voice (a voiceover over an interview).

The studies behind these numbers are TV documentaries, so a music-led ad may duck less on purpose. Record the first approved mix's real figures beside the flag lines. Also note the speech against the effects stem, which holds the shots' own sound as well as the designed effects; it has no flag line. A designed effect over a line is named from the sound list.

2. Mono fold-down

Sum left and right and compare its loudness with the stereo mix, for the whole mix and for the speech (steps below). Starting flags: more than 3 LU lost overall, or 1 LU in the speech. Watch the stereo correlation on a phase meter: a negative reading means something cancels in mono (steps below). Starting flags: the lowest reading under −0.5, or more than 1% of the time below 0.

3. The phone band

The phone band is a small-speaker stand-in, never a phone. Play the stems through a 300 Hz high-pass and an 8 kHz low-pass, each run twice for a steeper slope. Measure speech against the rest during speech, stretch by stretch as in check 1, with the other stems mixed into one for the rest. Flag a margin under 10 LU (steps below).

Then check the music's bass on a spectrum (steps below): on the music stem, over a passage where the bass plays, read the strongest peak between 40 and 200 Hz (the bass note) and its harmonics between 200 Hz and 1 kHz. Flag it when the harmonics sit more than 20 dB under the bass note: bass that is only sub disappears on a phone.

4. Planned quiet moments

Run each one through a playback-compression stand-in (see Sound design and mix → Planned silence and playback compression).

5. Your own phone and earbuds

On the first review render, listen once on your phone's speaker and once on earbuds. Write one line per device beside the flags, and compare them. Are the words clear, and does the bass still read?

Do it yourself

Measure loudness over one stretch of speech (ffmpeg), for check 1

  1. ffmpeg's ebur128 filter does it one stretch at a time, here 12.0 to 15.5 s of the music stem; its summary gives the integrated loudness:

ffmpeg -nostats -i stem_music.wav -af "atrim=12.0:15.5,ebur128" -f null -

  1. Run the same stretch on the speech stem and subtract. Repeat for each stretch of speech.

Measure the mono fold-down (ffmpeg), for check 2

  1. Sum left and right and measure the loudness (add atrim for a speech stretch, as above):

ffmpeg -nostats -i mix.wav -af "pan=stereo|c0=0.5*c0+0.5*c1|c1=0.5*c0+0.5*c1,ebur128" -f null -

  1. Measure the stereo mix's loudness as well (the command is in Sound design and mix → Loudness: −23 TV, −16 online, −14 short social → Do it yourself), and compare.

Read the stereo correlation (Audition or ffmpeg), for check 2

  1. Watch Audition's Phase Meter, or use ffmpeg's aphasemeter.

  2. This prints one reading per audio frame:

ffmpeg -i mix.wav -af aphasemeter=video=0,ametadata=print:key=lavfi.aphasemeter.phase -f null -

Measure through the phone band (ffmpeg), for check 3

  1. Measure the speech stem through the band, one stretch at a time:

ffmpeg -nostats -i stem_dialogue.wav -af "highpass=f=300,highpass=f=300,lowpass=f=8000,lowpass=f=8000,atrim=12.0:15.5,ebur128" -f null -

  1. Measure the rest the same way, with the other stems mixed into one:

ffmpeg -nostats -i stem_music.wav -i stem_effects.wav -i stem_ambience.wav -filter_complex "amix=inputs=3:normalize=0,highpass=f=300,highpass=f=300,lowpass=f=8000,lowpass=f=8000,atrim=12.0:15.5,ebur128" -f null -

  1. Subtract, stretch by stretch.

Read the bass on a spectrum (Audition), for check 3

  1. Open Audition's Frequency Analysis on the music stem, over a passage where the bass plays.

  2. Read the strongest peak between 40 and 200 Hz, then its harmonics between 200 Hz and 1 kHz.

Sources

bottom of page