Using Traktomizer

Stage 1: Uploading your Song

Traktomizer wants to compare your song with at least 16 benchmark songs but no more than 24. You do not need physical access to any benchmark songs or to know what they are. You only need to choose one Benchmark song and Traktomizer will find the rest. So let’s refer to these songs collectively as your Benchmark Analysis Pool.

Assuming you have successfully created an account at analyser.traktomizer.com, login and you’ll be presented with a screen which will eventually list all your analysed songs.

Click the Upload Song button to launch the upload pop-up.

Enter the Title of your song. (which is allowed to be different from the .mp3 file’s title)

Click the ‘choose file’ button and navigate to the .mp3 file on your computer you wish to analyse. Your audio upload must be a .mp3 file prepared as 44100k x 16bit x 128k (mono or stereo). Your file must not be larger than 8MB in size.

Now enter your Benchmark Artist and Song Title. Here you want to be thinking about a song/sound you aspire to for your song — which track really nails it for you? That’s your Benchmark. You can choose a different benchmark next time and so on. Your benchmark song should typically be in the same genre or at least one next to it.

Enter the Genre you’re going for in the last input box.

Click ‘Upload’.

That’s all you need to do to get the analyser rolling...

Stage 2: Processing your Song

As soon as you click Upload the dialog will close and the page will report Traktomizer’s progress from time to time.

Note that should Traktomizer not be able to find the required meta-data for the Benchmark artist/song you nominated, the process halts and requests you to start again with an alternative. The process will also fail if your upload file size is larger than the maximum allowed 8MB per song. Credits are not deducted if the analyser is unable to complete its work!

So what’s going on under the hood?

Traktomizer sets about its first order of business of analysing your song; measuring the individual Audio Features we care about. Then using your benchmark song as a seed for similar songs, Traktomizer identifies 16 to 24 more songs it deems appropriate for this analysis task.

Traktomizer chooses benchmarks using the Spotify API similarity search algorithm (the original version conceived and created by The EchoNest), which is normally used for the creation of user playlists. Songs may individually be characterised and differentiated by like-genre, like-mood, like-tempo, like-danceability and so on.

Once Traktomizer has settled upon 16 to 24 songs (from hundreds of candidates) it retrieves each audio feature vector value we are interested in. Then Traktomizer collates, aggregates and calculates thousands of deviations to formulate the measures by which it will rate your song’s overall chart-ability and each individual audio features power to influence it.

It takes Traktomizer around 60 seconds to retrieve the data it needs and compute all audio feature ranges of vectors. Typically these data points reflect the standard deviation of 18 songs altogether, perhaps a few more. (Your song + your Benchmark + 16 more chosen by Traktomizer).

Traktomizer will also retrieve this week's top 50 songs on Spotify as well as the top 50 songs on the current 7-Digital Pop Charts. Traktomizer retrieves these charts in search of 5-7 more similar songs and may therefore end up with a total of around 25 songs altogether. If your own song’s genre has no counterpart currently on the charts Traktomizer may apply discretionary weightings to give you maximum confidence in the final results.

Traktomizer will also have identified the top song of each benchmark artist by scrutinising the data contained in their top 10 songs; a double-check which ensures the benchmark pool as a whole represents what you're going for.

Stage 3: Audio Features - Analysis

Within a few minutes:

Traktomizer calculated around 20 audio features of your song and retrieved the same audio features based on existing analysis for each benchmark song (out of approx. 210 candidates).

Traktomizer scrutinised 50 currently charting songs to possibly add up to 5 more if it found 5 in the desired genre. The final Benchmark Analysis Pool should comprise your song (1) + the benchmark song + 16-24 similar songs + 5-7 similar charting songs. These songs are usually the most loved, the most recommended, the highest ranked and most similar to your original seed.

Traktomizer tallied, sorted and calculated thousands of intersections and comparisons, aggregations and sums before appraising each of your song's audio features relative to the benchmark pool top scorers and song presenting 1st, 2nd and 3rd ranking audio features, as well as the mean ranges of all audio features.

Traktomizer is trained on song data from Librosa's 1000 000 Song Data Set, plus the last 5 years of Billboard U.S. pop chart data, plus current Spotify meta and chart data and 7-Digital chart data. These milliard data points form a standard by which Traktomizer is able to compare your song to what is already known.

At first glance it may appear more desirable to achieve scores which are always higher than your Benchmark Analysis Pool values, but in reality sometimes lower scores may prove more desirable, Taking such factors into account Traktomizer is able to rate audio features in orders of magnitude, attributing points based on an index of weightings.

Certain audio features have been selected to score bonus points which are usually awarded when several highly influential audio features return positive measurements as a group.

Roughly 2-3 minutes from the start and Traktomizer will have produced beautifully animated HTML5 charts, presenting them in dashboard style views, complete with speedometer style gauges for quickly visualising your song's strengths and weaknesses.

Stage 4: Top Genre Analysis

While all this was going on Traktomizer spawned a second thread; another kind of similarity search based on different criteria from the first thread. This second search will yield around 50 more similar songs who’s audio features most closely match with your own song’s vector values. Traktomizer scrutinises every tag, keyword, genre and term from the entire batch to determine a top 15 list of genres appropriate for your song.

By around 4 minutes Traktomizer has produced a PDF Report listing the Song Title and Artist of each of the secondary 50 matching songs. Additionally all the words were reduced to a sorted list of the 15 most common terms and genres in which placement of your song is likely to achieve maximum discoverability and recommendations.

Scrutinising the Genres PDF you are likely discover that your song demonstrates potential popularity in various genres you have never considered or even heard of and mentioning these genres in your blogs, news and the wide variety of available social media will draw additional plays and new fans.

Stage 5: Charts & Reports

Traktomizer has by now produced 2 reports in PDF format:

The first is a Dashboard Audio Analysis Report presenting your song's Chartability Rating; your weakest 3 to 5 audio properties; and how well your song stacked up against the benchmark analysis pool (individually and collectively).

The report highlights strong and weak audio features drawing your attention to make changes there first (if you think you can improve them).

Making changes to these audio features first is being presented as the most likely action path for achieving the maximum overall improvement possible but with the least amount of change.

Even if other areas are also found wanting attention, whatever attention you give them will have lower orders of impact and influence on the end result.

The contents of this report may also be visualised with bar charts and gauges. These detailed visuals allow you to see how well your song performed across each audio feature for each benchmark song.

The second PDF contains the Top 15 Genres Analysis. These 15 terms are likely the most important to mention in your promotions, conversations, tweets, statuses and various tagging.

Traktomizer invoked another thread when it searched for all your benchmark songs’ 30 second mp3 previews on Spotify.com. It probably found previews for most if not all of the songs but its not guaranteed to find a preview for every song and may sometimes find the wrong preview despite best efforts. A playlist was created and pushed to your Analyser Song List so that you can hear what your song has been compared to — especially helpful if you don’t know some of the songs Traktomizer included in your Benchmark Analysis Pool.

Traktomizer also probed the NextBigSound.com API and using your artist name it first discovered which social media sites you are most active on and then tried to find the same statistics for each artist in your benchmark analysis pool. Tried, because that data is not always available for all artists. Finally an interactive spark-line chart, comparing your social activity with your benchmarks’ on the social sites you use has also been linked into your Analyser Song List.

Stage 7: Analyser Completion

When Traktomizer finishes its analysis tasks the web page will automatically refresh and your Analyser Song List will contain all the links to the reports, charts, previews and summary data described above. Traktomizer shows you all the strengths and weaknesses that count from a recommendabot's point of view. Some of these audio features will likely take the focus in your next production iteration. You can upload the new iteration for analysis and by so doing continuously check chartability and audio feature optimisation and improvement.

Consider Your Song Traktomized!