netik/sndchecker

Experimental code to see if it's possible to determine if a sound file is of "low quality" or "poor dynamic range."

sndchecker

Experimental code to see if it's possible to determine if a sound file is of "low quality" or "poor dynamic range."

Usage: ./sndchecker [options] filename

sndchecker returns a score indicating how many buckets in a file
conform to a specified loudness threshold.

  --threshold  power limit for a "good" bucket 0.0 to 1.0 [defaut: 0.13]
  --size       samples per bucket [defaut: 22000 / 500mS at 44.1Khz]
  --json       return score in json
  --histogram  return a URL with the bucket list for graphing
  --verbose    verbose logging and statistics

Algorithm

divide up the track into a number of buckets, around 500mS / bucket
for each bucket, calculate loudness for the entire bucket
  if loudness > threshold                                                                                     
     count bucket as good
     reset the per-bucket counter                                                                                

divide the total number of buckets by the "good" buckets.                                                 
  return this as a score between 0-100%

We consider anything < 40% to be a "bad track" with many quiet passages.

False Positives

• Songs with many changes in loudness
• Ambient, "Quiet" tracks
• Classical
• Non-maximized / loud tracks

Most Pop music works very well with this returning high scores. Old timey music, not so much.

Todo

• Research better music identification techniques