Some program material, such as found on BBC Radio 3, uses wide dynamic range (about 20 dB, as measured by EBU R-128 meter). To make this material more suitable for portable listening, the dynamic range should be reduced towards 10 dB. This project implements a few ffmpeg filters and further builds on them to provide a general-purpose automatic procedure for reducing dynamic range and transcoding for portable listening.
This is a port of qaac smart limiter. It should be applied to an audio signal which is already normalized and generally well-limited, except possibly for a few stray spikes. It then fixes those spikes in the most seamless manner.
This is a port of qaac compressor. It's a classic compressor (= with attack and release times, rooted in analog designs). The implementation is specifically based on Digital Dynamic Range Compressor Design - A Tutorial and Analysis, JAES2012. In addition, this implementation provides the delay (aka lookahead) control to cope better with sharp attacks.
Unlike classic compressors, mydrc doesn't do close envelope following
and rapid level changes. Instead, it uses a huge lookahead buffer to adjust
the volume gradually, over the period of a few seconds. The implementation
is based on LoRd_MuldeR's Dynamic Audio Normalizer
(ported to ffmpeg as dynaudnorm). It uses Gaussian smoothing filter.
Unlike dynaudnorm, however, it doesn't do upward compression and actually
doesn't try to normalize the volume; that is, it does not "even out" the volume
of quiet and loud sections completely. Instead, it uses downward compression
curve similar to that of qadrc, so that loud parts stay relatively loud,
and soft parts remain relatively soft.
This is the script which puts it all together. It checks to see
if the dynamic range of the input signal is wider than 10 dB, and sets up
parallel compression,
mixing the outputs of qadrc and mydrc. It also does automatic mono
detection and volume normalization.
This is the volume normalization part which can be used separately. It normalizes to the average of ReplayGain 1.0 and 2.0 specifications. ReplayGain 2.0 uses EBU R-128 algorithm with -18 LUFS reference level, and is generally considered to supersede ReplayGain 1.0. However, ReplayGain 1.0 is more sensitive to peak levels, which is important for portable use. The difference between ReplayGain 1.0 and 2.0, after applying compression, is often around 2 dB (that is, ReplayGain 2.0 suggests volume go 1 dB up, while ReplayGain 1.0 suggests volume go 1 dB down).
You can use apicker
to pick the right part of the program (= its start and end times)
before passing it down to transcode. The picture below shows the apicker
UI with qadrc and mydrc downward compression curves (the blue and the red
one respectively).
