AI Similarity Search

One massive headache of our job is having a great reference track but too large a catalog to quickly find a soundalike. Similarity Search instantly scans your DISCO for similar-sounding tracks, streamlining the search and review process with AI precision.

alternatetext
Chat icon
Transcript
CONTENTS

Introduction to Similarity Search

Similarity Search uses DISCO's AI technology to analyze and compare reference tracks, and YouTube audio, against all the tracks in your DISCO. Having access to an instant list of similar tracks accelerates the creative search and selection process.

In this lesson, you will learn:

  1. Two ways to utilize Similarity Search
  2. How to rate results and provide according feedback 
  3. Important notes to keep in mind when using the feature
  4. Frequently asked questions

How to Find Similar Tracks

Search With a YouTube Link

Simply enter a link in the Search Bar and the AI will return what it believes are the most similar tracks in your DISCO. The results will be ordered from the most to least similar.

Browse Through Your Catalog

In the track panel, click on the “Similar tracks” tab, and the most similar tracks in your DISCO will be returned.

Rate the Results

Development on Similarity Search is ongoing, and you can help us make it even sharper by providing feedback. When you see results you think are great (yay!) - or even not so great (boo!) - please let us know by clicking the thumbs up or down button on the suggested track. That’ll help train the AI, and ultimately make the search-to-pitch process breezier as the technology advances.

Important to Know:

  • Videos stored in your DISCO are not currently analyzed by Similarity Search. 
  • Similarity Search works better with larger catalogs. The more tracks available for analysis, the more accurate the resulting suggestions.
  • For Discovery Suite subscribers: Only tracks that are auto-tagged will be included in Similarity Search results.
  • For Supervisor plans: All newly uploaded and imported tracks are automatically added to Similarity Search.

FAQs:

How does Similarity Search work?

Using AI, we have developed our own Similarity Search that analyzes and compares audio across many different dimensions. Suffice it to say, there is a bit of magic under the hood that we won't go into detail about here. What we can say, however, is that our Similarity Search does not take tags, metadata, or any other high-level descriptive track information into account. Instead, it performs the analysis purely on and around the audio itself. 

Who can Access Similarity Search?

Similarity Search is currently in beta and available by request only for Discovery Suite subscribers and Music Supervisors. To request access, please get in touch with us at support@disco.ac. 

How long will Similarity Search be in beta? 

The length of the beta period is currently undefined and may change depending on user feedback and suggestion(s).

How accurate is Similarity Search?

We have done a lot of testing before releasing Similarity Search, and we believe it provides accurate results most of the time. However, there are a few factors that will impact the quality of the results, including:

  • The amount of tracks in your DISCO
  • The more tracks you have, the greater the likelihood of finding similar tracks. 
  • The amount of tracks you have auto-tagged as a Discovery Suite subscriber.
  • Only tracks that are auto-tagged have also been analyzed for the purpose of Similarity Search. To increase the quality of your results, make sure to auto-tag as many tracks as you can. 
  • The variety of your catalog
  • If, for example, most of your tracks are rock and punk, then finding similar tracks to a classical or soul song is unlikely (not impossible, but unlikely!). While we still show you results of what we deem most similar, the quality of these results may be lower.
  • The quality of the reference track
  • If you search with a YouTube video that is very low quality and/or somehow convoluted (for example, is an audio mashup or has added voiceover), the quality of results may be lower. Best practice is to search using videos with high-quality audio. 

I had a track auto-tagged - how long before I get similar track results?

This may vary a little bit, depending on the time of day. Generally speaking, results should start showing up within 5 to 10 minutes.

Can I search with just a section of a video or track?

Not yet. Similarity Search is currently always performed using the entire video or song. 

Can I filter or refine my search results?

Not yet, but we are working on it. 

Does Similarity Search work in DISCO Catalogs?

Not yet, but it is scheduled for development soon. 

I am seeing duplicate results. Why?

When you perform a search, you may find that some results get duplicated. At this point, we are showing all results, even if they are duplicates of one another. If you find duplicates, that means you have the same audio more than once in your DISCO. You can either ignore this, or choose to merge the tracks to keep your DISCO neat and tidy. Learn more about merging.

My YouTube search failed - What can I do?

If a search fails and no results are provided, it may be because of one of the following reasons:

  • Lack of auto-tagged tracks
  • Tracks get added to Similarity Search results through auto-tagging. If you haven't auto-tagged any tracks in your DISCO, you will not see any results when using Similarity Search.
  • Territory restrictions
  • Sometimes, video comes with territory restrictions and therefore cannot be accessed for analysis. There are a few of things you can try in this circumstance:
  1. Find another version of the song on YouTube and try again. 
  2. Find a similar cover of the song. If you happen to have the song as an MP3 already, upload it to DISCO and try it that way.
The very same restrictions may also lead to a situation where you search for a video in DISCO, and results display without play capabilities. Even though the video won't play in DISCO, the results can still be trusted.
  • Length of the video
  • Ensure that videos are no longer than 30 minutes.
Failed search warning