Establish a common pattern for post-gain

[iver56]

A lot of the transforms may make the sound louder, which may result in digital clipping down the line. We should have a common way of doing post-gain to deal with this issue. Or a class that does the post-gain.

Like

GainCompensation(transform, method)

Examples:

GainCompensation(AddGaussianNoise(), method="same_rms")
GainCompensation(AddGaussianNoise(), method="same_lufs")
GainCompensation(AddGaussianNoise(), method="peak_normalize_always")
GainCompensation(AddGaussianNoise(), method="peak_normalize_if_too_loud")
GainCompensation(AddGaussianNoise(), method="target_rms", target_rms=0.01)
GainCompensation(AddGaussianNoise(), method="target_lufs", target_lufs=-23.0)
GainCompensation(AddGaussianNoise(), method="target_peak_dbfs", target_peak_dbfs=-1.0)
GainCompensation(AddGaussianNoise(), method="target_true_peak_dbfs", target_true_peak_dbfs=-1.0)  # not to be supported in initial implementation

[mmxgn]

I like the idea of a GainCompensation transform.

[iver56]

Or maybe PostGain is a better name?

[mmxgn]

That works too. Alternative names: AudioBus or Fader

[iver56]

[iver56]

I think I prefer PostGain. Because

• GainCompensation suggests that it's always about compensation, but not all methods I proposed are always "compensation"
• AudioBus sounds a bit too vague for me. I wouldn't understand what it did by just looking at the name

[iver56]

What's better?

• target_rms
• target_rms_db (and give the RMS value in decibel)