Interview With Dr. Jan L. van Os


Introduction

Below is an interview with Dr. Jan L. van Os, inventor of the MelodyCatcher, a music search engine for finding songs by melody line. For more information, or to try out the MelodyCatcher, please visit his website at melodycatcher.com.

This interview was conducted by email from Oct. 20-Nov.10, 2009.

Interview

Amy Gold: What is the MelodyCatcher?

Dr. Jan L. van Os: The MelodyCatcher is a music search engine in which you can enter a melody or musical line via a simple keyboard-like Java applet interface. The melody is then searched in our database and the closest matches, if any, are returned. Each search result is presented with a link to the original audio file along with its source on the Internet, and a play button for instant playback. Although songs can also be searched using text, what makes the MelodyCatcher different from conventional search engines is that it is not limited to just text. The MelodyCatcher can be highly useful to those searching for songs for which little or no textual data are available, e.g., the song is an instrumental, or the user does not remember the artist, title, or any lyrics samples.

AG: Tell us a bit about how you came up with the idea for the MelodyCatcher.

JLVO: For me, it all started almost sixty years ago (I am now 77) when I was a veterinary student in Holland and started playing Gypsy music as a hobby, which I still do today. You can watch me playing the cimbalom (with the hammers) at YouTube.

When I retired, I started to make a collection of Hungarian melodies from which I made some 4000 MIDI files along with the related demos of MP3 files I collected. These files are all on cimbalom.nl (my other web site). Currently, cimbalom.nl has the largest on-line collection of Hungarian Folk. These days, I get several new melodies from Hungary from visitors to cimbalom.nl almost every week ... a small world.

To build this collection, I needed a melody finder that could verify that each new melody to be added to the collection was not already included under a different name to minimize the number of duplicates, since many melodies have more than one title. This is how the MelodyCatcher (MC) started. First, I came up with an optimal matching contour algorithm. Then, I needed someone to write the software and through his virtual piano keyboard software, I found Florian Bomers. Today, looking at the user-friendly tool he developed, I have been lucky with all his skills that he brought to this project. In fact, I worked together with him by email for some 6 years, before ever meeting him in person (modern world).

AG: How does the MelodyCatcher work?

JLVO: The MC uses an automated web crawler to search the Internet for publicly available melodies. So far, mostly MIDI files and ringtones are indexed, but we plan to extend the system to also find other file types. The crawler adds these melodies to our database using our patent pending "contour" algorithm, which allows for accurate, yet error-tolerant searches for those using our web site (melodycatcher.com). So far, we have crawled some 40 specific web sites (that is, sites with melodies with a recognizable melodic contour), yielding already more than 60,000 melodies for melody search. After expanding the infrastructure, we can run the crawler 24/7, which means the melody database will grow to eventually include all suitable melodies on the net. For the necessary funding to do this, however, we're looking for an investor.

Because of its refined matching algorithm, the MC is far more accurate than other melody finders, because most of them are based upon complex techniques to recognize a melody from polyphonic music, so far with less accurate results. Other sites use "audio finger printing," a technique to identify recordings. This technique is accurate, but limited to the original records in the database, and of no use for melodies recorded in another way.

AG: I have played around with the MelodyCatcher a bit and it is really nice. I was able to find a lot of classical music themes with ease, as well as many Beatles and Elvis tunes. Besides classical music, pop, and Hungarian folk music, what other types and genres of music are included in your database?

JLVO: As mentioned previously, the ultimate goal is to include all suitable melodies on the net. The current database is just a reflection of the sites that we so far have chosen to crawl. In doing so, we have not focused on classical music. Other genres include pop (in a broad sense), folk, evergreens, hymns, movie soundtracks, TV themes, musicals and Christmas carols.

AG: For oldies music fans, about how many pop songs from the 1950s, 1960s, and 1970s are currently in your database?

JLVO: Honestly, there is no exact answer to that question, since the web sites that we index usually do not include information on the genre or time period - at least not in a way that is obvious to our web crawler. But I guess that at least some thousands of them are available for melody search, obviously with overlap for the better known songs. For text search, there are many more to be found.

As such, this is an interesting subject, since most of the numerous music search engines are mostly focusing on mainstream music of today’s performers. At best, just the most well known old melodies are included in their database. We have tried to also include more of the lesser known old melodies. Honestly, if you ask me, many of these almost forgotten oldies are better melodies as such than lots of music popular today. But then my impression might also have to do with the good old memories of my generation.

AG: When entering the melody, do you need to worry about the original key your song is in, or does the MelodyCatcher transpose melodies automatically?

JLVO: The MC allows you to enter your melody in any key. This is made possible by the use of a highly refined melodic contour we developed that is based in part on the Parsons contour which, simply put, looks only at the three directions of succeeding notes of a melody. Such a contour remains the same for any key in which a melody is played.

In practice, users more familiar with playing notes might prefer to start in C for melodies in a major key or in A when minor. But in fact, a user can start with any note, as long as he/she is capable of hearing, either when playing or replaying the input, whether it actually is the wanted melody. The interface allows you to easily correct wrong notes.

AG: Can you tell us some more about the process of playing and searching for a wanted melody?

JLVO: In general, at most some 10 of the first notes of a melody have to be played, to find in the top of the search results those melodies for which the systems crawler has found a suitable contour on the Internet. All played notes are instantly displayed in two ways: as notes in a graphical contour and as text in a note name display field. A melody like "Happy Birthday", when played in C, is shown as: G2 G2 A2 G2 C3 B2 G2 G2 A2 G2 D3 C3. For that matter, the text of these notes can be copied and pasted. This offers an easy way to email a melody as text to a friend who then can paste this text in the MelodyCatcher to play it back.

Before searching each input can be played and when needed, corrected in different ways:

- by grabbing with the mouse a note of the graphical contour and dragging it up or down to alter the pitch of the note
- by removing, adding or changing the text of one or more notes

By default, it is assumed that users enter the beginning of a melody. Any part of the wanted melody, however, can be entered by using the "Search Anywhere" option. This can be useful if:

- a user is uncertain about the first notes of a melody
- the beginning of the wanted melody is less distinctive
- a typical part of a melody is best remembered

AG: What are your future plans for the MC?

JLVO: We have a lot of plans! We want to continue our efforts on the crawler, so that it makes more types of audio files accessible with the MC, e.g. suitable polyphonic MIDI files. We also want to provide more information in the search results, gathered from meta data in audio files and from the web page containing the audio file, so that the database will grow further. We want to carefully extend the database by crawling selected web sites. And as mentioned previously, we're also looking for a partner or sponsor to invest in, or integrate into, a bigger infrastructure for the server so that the crawler can run 24/7 and eventually index the entire Internet!

AG: Dr. van Os, it was a real pleasure interviewing you. Thank you so much for describing your wonderful invention and for making it available to oldies and other music lovers around the world.

JLVO: You're welcome. We are pleased with your interest in our tool and I enjoyed answering your questions.