Using ffmpeg to split audio files to the size expected by ffmpeg

split-audio-with-ffmpeg-for-second-life.md

Instructions on how to use ffmpeg

To split sounds for usage in Second Life.

• Gwyneth Llewelyn (20220608)

Install ffmpeg and LAME

Your mileage may wary.

Convert MP3 to WAV

ffmpeg doesn’t work natively with MP3, it uses an external plugin for that. Therefore, we’ll first convert it to WAV.

ffmpeg -i Sugar\ Plum\ Dark\ Mix.mp3 sugar-plum-dark-mix.wav

If you wish, you can check with ffprobe before to see the average bitrate. Our example: 320 kb/s, so we do:

ffmpeg -i Sugar\ Plum\ Dark\ Mix.mp3 -b:a 320k -minrate 256k sugar-plum-dark-mix.wav

Turn volume louder

This requires several passes.

First, determine the average and peak loudness:

ffmpeg -i sugar-plum-dark-mix.wav -filter:a volumedetect -f null /dev/null

This should show something like:

[Parsed_volumedetect_0 @ 0x7fcb8400e1c0] mean_volume: -24.5 dB
[Parsed_volumedetect_0 @ 0x7fcb8400e1c0] max_volume: -0.9 dB

Better still, get it working with loudnorm:

ffmpeg -i sugar-plum-dark-mix.wav -af loudnorm=I=-16:TP=-1.5:LRA=11:print_format=json -f null -

which results in:

{
        "input_i" : "-23.21",
        "input_tp" : "-0.87",
        "input_lra" : "12.50",
        "input_thresh" : "-33.81",
        "output_i" : "-14.47",
        "output_tp" : "-1.50",
        "output_lra" : "5.40",
        "output_thresh" : "-24.73",
        "normalization_type" : "dynamic",
        "target_offset" : "-1.53"
}

Note that mean_volume is equivalent to input_iwhile max_volume comes close to input_tp. In the above scenario, clearly -16dB is not enough to get this to acceptable levels; so I’ve tried to go even higher than that:

ffmpeg -i sugar-plum-dark-mix.wav -af loudnorm=I=-8:TP=0:LRA=12.5:print_format=json -f null -

Radical!

{
        "input_i" : "-23.21",
        "input_tp" : "-0.87",
        "input_lra" : "12.50",
        "input_thresh" : "-33.81",
        "output_i" : "-11.23",
        "output_tp" : "+0.00",
        "output_lra" : "5.40",
        "output_thresh" : "-21.44",
        "normalization_type" : "dynamic",
        "target_offset" : "3.23"
}

What we need now is to do a second pass, feeding those values into the call to the loudnorm filter:

ffmpeg -i sugar-plum-dark-mix.wav -af loudnorm=I=-8:TP=-0:LRA=12.5:measured_I=-23.21:measured_TP=-0.87:measured_LRA=12.50:measured_thresh=-33.81:offset=3.23:linear=true:print_format=summary -ar 44100 sugar-plum-dark-mix-louder.wav

-ar 44100 may not be necessary.

Result:

Input Integrated:    -23.2 LUFS
Input True Peak:      -0.9 dBTP
Input LRA:            12.5 LU
Input Threshold:     -33.8 LUFS

Output Integrated:   -14.7 LUFS
Output True Peak:     +0.0 dBTP
Output LRA:            8.4 LU
Output Threshold:    -25.0 LUFS

Normalization Type:   Dynamic
Target Offset:        +6.7 LU

If there is garbage at the end, trim it:

ffmpeg -ss 00:00:00 -t 00:02:00.00 -i sugar-plum-dark-mix-louder.wav sugar-plum-dark-mix-louder-trimmed.wav

Turn stereo to mono

This should do the trick:

ffmpeg -i sugar-plum-dark-mix-louder-trimmed.wav -ac 1 sugar-plum-dark-mix-mono.wav

Note: bitrate should now be half of the stereo sound.

Chop it into manageable bits

ffmpeg -i sugar-plum-dark-mix-mono.wav -codec:a copy -vn -f segment -segment_time 9.9 -segment_start_number 1 split/%02d-sugar-plum-dark-mix.wav

Note that the %02dtemplate parameter will use 2 digits, so that segments are properly ordered when using ls -la.