From time to time we hear from a station that automatic music recognition is generally working well but for a certain program it performed very badly. We’ve investigated a number of these cases to try to reveal the causes of the problem. This FAQ describes some of the things we’ve found.
First we check that systems were operating properly at the time. We can look at logs in Spinitron and in ACRCloud, our partner who provides the recognition service. If the stream is not working at all then of course recognition won’t work. Sometimes a stream is unstable for a while leading to intermittent recognition failure. If there’s no sign of such trouble then…
Second, is the music played in ACRCloud’s database of acoustic fingerprints? If not then we don’t expect recognition to work and this can be specific to certain programs. As a rule of thumb, if the music is distributed for digital sales and streaming then it’s likely to be in the database. Music that was never reissued in any digital format is the kind of thing that might not be in the database.
Third, if the music is in ACRCloud’s database and the systems were working at the time then something else prevented recognition. We have found a number of things can degrade the audio in such a way as to make recognition unreliable or fail.
-
Turntables with a pitch adjustment feature. Good quality DJ tables have this feature. Don’t use it if you want recognition to work. If the table has quartz strobe speed indicator, use it to check the adjustment is set to zero.
-
Turntables without pitch adjustment feature. Many belt drive turntables, even audiophile ones, have inaccurate speed. So long as the speed is reasonably stable it doesn’t affect the listening experience for most users since a few percent is seldom noticed. If you’re using a table without its own strobe speed indicator, get some measurement gear to check or use a different table that does. Search on youtube for techniques and equipment for measuring turntable speed. We’re interested mainly accurate measurement of its absolute speed. Wow and flutter measure speed stability but recognition can fail if speed is very stable but just a percent off.
-
CD players with pitch adjustment. Some pro CD decks have a pitch adjustment, often hidden away in a menu that users seldom see. In one case we had a station that found one of their three CD players had a pitch set less than 1% off and that was enough to defeat recognition.
-
Digital playback with pitch adjustment or time stretch. Deejay apps that play audio files on a computer also offer both pitch adjustment and time stretch. These are very useful for certain performance tricks but aren’t compatible with recognition. Check that these features are turned off.
-
Pre-produced programs that have been adjusted for time. Sometimes we find that a station wants pre-produced programs to be a certain length, pretty much to the second, e.g. 58:00. This can be accomplished in a number of ways. One is to carefully deejay the show accomplishing this with adjusting segues and talk breaks. Another is to do it approximately and then adjust the completed audio file’s duration, either with pitch shift or time stretch. Either pitch shift or time stretch will affect recognition so please don’t use this technique.
-
Audio chain problems. We resolved one case at a station where a certain DJ’s pre-produced shows had poor recognition while it worked for everything else. The station manager was able to work with the DJ to resolve a technical issue in the DJ’s production at home. I don’t recall what the issue was in that case but they resolved it. I can imagine that improper gain staging that causes clipping would affect recognition.
-
Source file issues. We had another case in which certain audio files in an automation system were not recognized and it turned out that the files, which had an unknown history, some had a pitch shift (I can’t imagine how. Perhaps it was a vinyl rip?) and others had a sound quality a lot different from the CD.