Remove FFmpeg normalization causing "quiet" songs.

[phcreery]

Title

Audio downloaded with spotdl is relatively quiet.

Description

Audio normalization flags in the FFmpeg command reduces actual song loudness by almost 50%

Motivation and Context

dr.boaf on discord originally voiced the issue and I too took notice of it. Removing the normalization flags causes the converted audio to match closer to the actual song. The normalization causes the song to be very dull and can be very apparent in certain conditions. I believe @MikhailZex originally introduced the flags. Could you expand on the usage and purpose?

Testing / Screenshots (if appropriate)

Subject: Tristam - Over the Edge

The absence of apad_dir also makes the song the exact same length

Types of Changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)

Checklist

• My code follows the code style of this project
• My change requires a change to the documentation
• I have updated the documentation accordingly
• I have read the CONTRIBUTING document
• I have read the CORE VALUES document
• I have added tests to cover my changes
• All new and existing tests passed

[phcreery]

Just fixed a flag that increased bitrate. See https://trac.ffmpeg.org/wiki/Encode/MP3

before/after

[ghost]

The dullness is due to the dynaudnorm flag that causes local amplitude modulation instead of global amplitude modulation. This ensures consistent loudness with almost no variation - variation that's supposed to be an integral part of a song.

It has been removed multiple times in the past and then been reintroduced again and again. Kind of gave up on getting it out.

Try conversion of the song with only loudnorm, if you still feel the audio is dull, go ahead with this PR as is, else remove all flags except loudnorm.

You can checkout my personal modified spotdl copy with alternate flags here.

[xnetcat]

Please pull changes from upstream/dev

[phcreery]

My question is why is the audio modification in there at all. I feel like most people would want the unmodified audio.

[ghost]

Instrumental music on YouTube has wayyy to much loudness variation. Without normalisation, your have to change the volume on a song to song basis. This effect is far less pronounced for mainstream music.

P.S. It's mostly for that reason that Spotify itself implemented loudness normalisation in songs by default.

[phcreery]

That makes sense. I will mess with the normalization to see if I could improve it. For now, though, I feel it would be better to have inconsistently varying songs rather than a quieted song (in place over an already compressed mp3 file). Most older songs were naturally quieter.

[phcreery]

I feel like the dynaudnorm should be left out altogether since it applies normalization dynamically over windowed portions of the file and changed characteristics of the song. This is not desired since some songs intend areas to be quiet while others to be loud.

[ghost]

I feel like the dynaudnorm should be left out altogether since it applies normalization dynamically over windowed portions of the file and changed characteristics of the song. This is not desired since some songs intend areas to be quiet while others to be loud.

The dullness is due to the dynaudnorm flag that causes local amplitude modulation instead of global amplitude modulation. This ensures consistent loudness with almost no variation - variation that's supposed to be an integral part of a song.

It has been removed multiple times in the past and then been reintroduced again and again. Kind of gave up on getting it out.

Try conversion of the song with only loudnorm, if you still feel the audio is dull, go ahead with this PR as is, else remove all flags except loudnorm.

You can checkout my personal modified spotdl copy with alternate flags here.

[ghost]

From my set of experimentation, one other thing of note is setting a relatively high level for loudnorm (eg. loudnorm=-3) will cause "pumping" - rhythmic loudness-softness loops.

[ghost]

LGTM for the most part.

[xnetcat]

LGTM 👍

[ghost]

@Silverarmor, I think this PR is a Go.