Quantifying sound quality

#1

Is there a standard way of quantifying sound quality for the purpose of comparing two versions of the same recording? This would be particularly interesting in cases where the bitrate and file size are unreliable or downright meaningless because of subsequent reencoding of low quality source files.

I’ve seen people use spectrograms to determine sound quality visually, but I’m wondering if it could be boiled down to a number or a series or numbers for systematic/automated comparisons.

If not, a tool that facilitates manual side-by-side comparisons would also be useful.

#2

Interesting; I haven’t heard of anything like this! I have never been in the game myself, but I have gotten the impression from people who do this sort of thing that spectrograph analysis cannot be automated, for whatever reason. (I seriously don’t know why.) But people swear by their ability to look at a spectrum and make an educated guess about its quality.

1 Like
#3

I’ve been using this to get the true bitrate of files. https://github.com/dvorapa/true-bitrate

It’s using a simpler mechanic than looking at the spectrographs but it seems to work.

#4

Bitrate sounds useful, but I’d suggest looking at Dynamic Range as an additional option for gauging recording quality. It’s particularly useful for comparing quality of different recordings for a given album. See http://dr.loudness-war.info/, ex. http://dr.loudness-war.info/album/list?artist=Radiohead&album=OK+Computer. There are a few tools on github for calculating it, but I haven’t messed with them too much as of yet.

1 Like
#5

Not an expert, but parroting the people who act like they know what they’re talking about.

Quality cannot be quantified because there are different measurements. I’d say any one measurement can be quantified, but how they are weighted into what sounds “better” is subjective. Some (misinformed) people say bitrate is king, and already in this thread you see dynamic range as another important factor that has nothing to do with bitrate.

Other things that affect perceived quality can be noise/scratches (vinyl sources) and the master itself. Different remasters of the same album can sound different because of decisions the remastering employee(s) made.

You may have a somewhat hard time finding scientific research or commercial interest in the topic, because it’s already “solved”. 99+% of people are totally happy with 320kbps MP3, and even CD-quality FLAC is cheap enough to store. Services like Spotify default to a “normal” quality around 96 kbps. There might be viable researched published around 1980-2010 or so, when storage and bandwidth was more expensive.

What is still costly is video delivery, so I’d turn to papers published by Netflix and others to see the state of the art in subjective quality evaluation. Ex. here. And hope that audio is mentioned somewhere.

The best sources for more info, that I know of, are HydrogenAudio and Steve Hoffman music forums. Audiophile communities are plagued by pseudoscience that these two places seem pretty good at avoiding. Maybe you can ask there.

Also, I wonder if it may be less effort to just secure new copies of the files in question.

1 Like