Similarity Forum

General Category => General => Topic started by: Tatiana on February 21, 2013, 12:03:50

Title: accuracy of comparison
Post by: Tatiana on February 21, 2013, 12:03:50
I have recently begun to work with Similarity programm an did not find information about the accuracy of comparison of audiofiles.
Please, give your comments on the next question: what value of % of precise means that files are identical? If the precise value is near 80% can I be sure the files are the same?
Title: Re: accuracy of comparison
Post by: LordYanaek on February 22, 2013, 09:08:47
Until an official answers you, i can share my personal findings.
When using precise i set detection to 80% and auto-mark to 95% I've had "instrumental" and "normal" versions of tracks show as 89% similar for example. Well, the music is of course the same, but i don't want to remove either. Same for live and album tracks. It probably depend on your collection but i wouldn't trust anything below 90% for auto-mark.
Title: Re: accuracy of comparison
Post by: Admin on February 22, 2013, 12:28:37
Precise algorithm is most advanced algorithm and accurate, values > 95% shows in most cases what files are same, but be caution Similarity only scans first 1min of song, if it different in further block it just don't know about this. Most music files 1,5-3min duration and don't have such limitations.
Title: Re: accuracy of comparison
Post by: hsei on February 22, 2013, 16:15:04
Another note: You can set time length as additional criteria. To allow some inaccuracy in cuts at ends, set limit to 95 to 98% (depends on material). Setting this parameter also reduces compare time substantially, since files with unequal times (not the same as size) are disregarded and not compared by further algorithms. This circumvents also the problem of files which are similar at the first minute. Another advantage ist that typically different recordings (normal, instrumental, live) of the same song differ enough in time and thus are sorted out beforehand.