With this post I'm closing the series about my album Constant Change. There is a lot more I could report about. Lessons I learned, ideas I had. I'm trying to stick to two topics: free software and free music.
The whole project was conducted using free software. There are too many programs involved to mention all of them. To name a few outstanding ones: Ardour was used as Digital Audio Workstation. If you're interested in Linux and Recording, please do donate. The only sequencer in use was Seq24 which is still not fully stable but still very simple to use. The cover artwork was done entirely with Open Office. It may be not the finest one of the DTP tools but it does the job. All background pictures of sleeve, cover, and label were taken using a horribly cheap China-made Lomo-style camera made from plastics. This camera takes four pictures in one second and the outcome tends to be rather experimental.
The plan to release this album under the terms of a Creative Commons (CC) license came up before it was even started. It should be made using free software and it should be free. Later on I found that this enabled me to use many great samples from Freesound. Does CC licensed music keep artists from having success? I don't think so. But you can make the best music in the world, or the worst. It will not be a primary factor for your success. There seem to be only a few listeners choosing their music according to how good or bad it is. I think most people don't even consider what they like and what they don't. At least with a certain tendency they like what is advertised the best. And that's when record companies come into play. And that's where CC licensed music drops out.
Myspace and last.fm profiles won't help artists with this respect. Who uses the buggy CCHits? There is virtually no radio play for CC licensed music. And often times those artists create music they like and don't care about the masses. (Again that's when record companies... yeah.) So in case you're a free music advertisement specialist, get in touch.
Meanwhile I'll make some plans for my next project.
Still my time is rather busy and muse for new music is rare. So here we go with another post in the series about my album Constant Change. Abstract Factory was, unlike most other pieces on the album, not created with a particular plan in mind. The title originates from the field of pattern-based software architecture. For me it has more meaning than that. The track starts off with a concrete factory, merging into a constantly flowing situation with a melody line inspired by some Tangerine Dream piece which I'm unable to locate in my collection.
In our western world, many people work in some kind of abstract factory. There are no concrete products one can touch. There are no raw materials that become a product. Instead, we're hacking into our computers. Even people in companies producing real products there are large management departments that deal with nothing »real«. Even money has become unreal. From golden coins to bytes in a database. People working in software companies are not only using abstract factories, they are also working for such factories in the other sense. A long time before virtual reality became a major buzzword in the media and companies like Second Life started most of us have gone a long way from scraping together the items of basic necessity to some task that is on another level of reality.
Then there are those rare days when there is power outage and you're sitting there with your candle wondering how to cook the things that might get moldy in the fridge if the power doesn't come back soon. Perhaps you live somewhere on a hill and you watch the city lying there in the darkness with the cars being the only thing to dispense some light.
Maybe your portable MP3-Player has a good battery by then so that you can listen to Abstract Factory to support this feeling of surreality.
The last weeks have been particularly busy. As initially intended for a Sunday, here's another post about Constant Change, another one in the series. Wahnsinn v.2 is part of an aftershock or back flash from civilian service. It is all but a piece of easy listening music. Inspired by some of the sounds from the ZynAddSubFX software synthesizer I quickly came up with an experiment: »See how bad you can make it!« This piece is not meant to please. It is meant to be disturbing and scary.
After ten months of collaborating with mentally diseased persons I thought that I had seen a lot. But that simply was not true. I met a friend I hadn't seen for a while on a local rock festival and he kept telling me that he learned lipreading. »See the people there in the distance? I can see what they're saying!« He had become schizophrenic. Another good friend by then just had gone the way of being manic-depressive. In the beginning I had to kick his ass every time we met just to get him going for a beer or something. In the end my voice didn't reach his brains any more. But he saw God speaking to him from a cloud. Schizophrenic, manic-depressive: for outsiders these are just psychiatric terms. If you're a friend it is nothing you can understand. I would say you can only understand those states of mind if you're affected yourself. And as an affiliated it gets the worst it can get because you can't let go. You can't close the door behind you. Wahnsinn v.2 immediately made a connection to all of this in my head.
This piece is about going insane in a serious way. Don't do it. Only 25% of all manic-depressive people do not relapse at a later point in their life. Most schizophrenic persons never again reach a state of mind the others call normal.
And here we go with yet another post in the series about Constant Change. As »v.2« indicates Das Meer is a follower of an older piece by the same name. Its roots date back to the days during civilian service and the first sampling experience. Originally the name was motivated by a German movie from which some vocals samples and text passages were used. For the second version, those were of course replaced by our own material to conform with the requirements of the creative commons license. A few weeks after the original recording had been done live in a session I traveled to the Netherlands with my plenty old Fiat car and ended up at the beach. In the road I often listened to the original tape with version 1 of the song. Some of the impressions from that days were re-introduced into version 2.
Dunes near Schlagen, Netherlands
Have you ever thought about flushing yourself down the john? I must admit for some this might not be a pleasant imagination, especially if you just have watched Trainspotting. Anyway: for me it is an analogon for disappearing unanticipatedly, entering a new, more underground system. So that's what happened, I flushed myself down to the sea. Me and myself and the car had a trip to somewhere, open ended. Self-discovery works better when you're not in the usual place. The meditative basic setting of the piece combined with an unsettled lead organ resembles all of this well.
Vocals and the lead organ were performed by Benjamin Strohmaier, a friend and musician I am glad to have the opportunity to play with in my blues band.
This post is yet another one of the series about my album Constant Change. OneStepForward is a three-part piece. It can also be regarded as an intro, interlude, and outro.
A friend remarked that the piece could fit to a movie scene: somebody is in a sad mood and driving alone in their car. I like this description. One Step Forward was inspired by the idea of traveling from one situation in one's life to the next one. It is a joy to go ahead, to encounter new experiences, to get to know amiable people. But it is always is a farewell to what has been before. And then there comes this autobiographical component again.
One Step Forward parts one, two, and three represent steps to take - or that have been taken.
This post is another one in the series about my album Constant Change. The origins track Ten Long Months date back to the year 2002. By that time I was doing alternative civilian service at a factory. The factory actually was printing press that employed mentally diseased and disordered persons.
Interior of an Elka E9 home organ
that was sampled for Ten Long Months.
To cut a long story short: I liked the job when it came to working together with my »special colleagues«. But often times I was just doing the job of an unskilled worker instead. I was paid approx. 2,50€ per hour, including expenses. For every single day of being ill I had to bring a doctor's note. I was told that if I just don't come to work I will go to jail for up to two years. I had to convince the boss by citing the German constitution that I cannot be obliged to attend the prayer. It is a sad feeling when you experience your state exercising its power like this.
For the first time in my life a had a regular income. It wasn't much. But it was enough to buy instruments. I bought quite a few. And I sold quite of few. Ebay were happy with me. And then I came across the Smurf Soundfont sample editor and I was fascinated instantly by sampling. During civilian service I realized the first time that I had this urge for making music as a counterbalance to the rest. Although I had been using a Juno 106 synthesizer for quite some time, this was when I started to get some understanding of the technology involved in electronic music.
One day I found myself walking through the factory with two mics and a MiniDisk recorder. The boss gave me a doubtful look but I ignored it. I recorded many of the machines. A Heidelberg Tigel universal printing machine. A stapler used for stapling magazines. The lifting ramp. A cutting machine. And some of the countless waste paper baskets abused as drums.
What you hear as the drums in Ten Long Months are those machines I recorded, plus some of the waste paper baskets. The samples were sleeping on my hard disk since 2002. Then I loaded them into my Yamaha A4000 sampler and made a piece of music out of what I had.
The style in a certain way carries on my experiments with Smurf and a Korg M1 that I did back in those days.
This post is the first one of a series about the album Constant Change by DrNI, who is also the author of this weblog. I was asked by several persons to explain how I came to create the flute melodies in the track Computational Linguistics. Apart from that, I feel the urge to explain certain things about the album that are not so obvious to the listener. The language of this series will be English as I aim to address a broader audience than I usually do in this blog.
Introduction: Machine-Created Melodies
First of all, what most listeners might not have noticed is that the flute melodies in Computational Linguistics were neither composed (like most background parts) nor improvised (like most lead parts) in the other tracks. They were generated from linguistic data. For this purpose I wrote a specialized software that operates on syntactic analyses of language and converts it to music. The aim of this first post in the series is to explain why and how I did this. I hope I will be able to tell this story in a convenient way. After all, I am studying computational linguistics for more than three years now and it sometimes is hard to get down and back to the common ground of other people.
Motivation and Science vs. Arts
In many if not in all fields of science, people publishing have to give reasons. They need to state what they want to find out and they need to say why. This is also true for computational linguistics. No academic reader will take you for serious if you just do something without stating why you do it. In art this may be different. I am not the person to judge about this. However, one of the views may be that art is to be free of a purpose. The only purpose of art is to be art. Otherwise it would be design. Then again, if you ask the artist about his or her work you may find that he or she has a clear intention of what this or that work should say or he or she might have had a certain reason to do things the way they were done.
My plan was to create a piece about computational linguistics (CL). Obviously, CL is about to become my profession, if it has not become it already. Apart from that, the beginning of my studies in CL have been one of the milestones of my life. As Constant Change has an autobiographical component there was no way around the topic. It must have been in early 2006 during a course at the University when I first had the idea to convert linguistic tree structures into melodies. Music is structure-based and created by humans and so is language. Linguistic trees are one way of many of structuring human language. Will the translation of this representation of what the human mind produces carry over to something that is acceptable as a human-created structure in music? This is the question that led to the experiment.
Linguistic Tree Structures
To readers being familiar with linguistics this is not a new story. So here we go with a short excursion for the default audience. This section is about syntactic tree structures which are the most common way of analyzing the structure of a sentence. This field of linguistics is called syntax. Syntacticians try to form models that describe human language. Those models describe how words can be put together to form sentences. They usually do not make statements about how language works in the brain. There are various theories and most of them use the linguistic tree as a graphic representation. The following figure shows a part of a linguistic tree (click to see all of it).
Tree view of sentence s1 from the BROWN Sampler (part of the Penn Treebank)
as shipped with TIGERSearch (click to see the full tree).
Note that this figure is not distributed under the CC license terms named
in the right hand side bar.
Note that the on the bottom level of the figure there are the words of the sentence plus some information on what kinds of words they are. E.g. DT stands for determiner and NN for normal noun. Above this so called word level there is the syntactic level consisting of a tree of connected nodes. Each node is assigned a type. There are noun phrases (NP), prepositional phrases (PP) and the like. A phrase is not a whole sentence. It is rather due to the syntax theory applied to the sentence than due to interpunctuation what makes a phrase.
The example above is taken from the BROWN Sampler which essentially is a part of the Penn Treebank. A treebank is a (large) collection of sentences plus the corresponding syntactic trees. The BROWN sampler also provided the data that were eventually used in the piece of music.
Feature Transfer
The key point of creating music from linguistic trees is feature transfer. My approach works on the word level. This is to say that the program processes the input sentences word by word. Each word has a number of features. In the example above, the features are (apart from the word itself) the type of word (part of speech) and the length of the path you have to travel until you hit the top node (or root) in the tree. For my approach the phrase types themselves were ignored. Other treebanks include morphological information such as case, number, and gender. These are also irrelevant to the approach.
There is another thing that has features: A note in music. Notes have primarily the features of pitch and duration. Volume is also a feature. People studying music will find this definition of a note very imprecise. However, I had to narrow down features to a concrete level. There is no doubt that a note may have many other features, including the actual tone eventually produced. For now, let us see notes as pitch, duration and volume.
The program I wrote is very generic. It is capable of mapping the features part of speech and path length from the word to the features pitch duration and volume of the note. The exact way of mapping is specified externally in a control file.
Concrete Application and Results
For the track Computational Linguistics I used the path length for the pitch of the notes. The longer the path the higher the pitch. However, pitch is not just set note by note in half tones. It is mapped to a given scale. In the piece this scale is E minor. This helps to ensure that the output is not complete noise. Part of speech information was mapped to duration and volume of the note. E.g. verbs are louder and longer than determiners or adjectives. Punctuation results into pause. Note that this is only one of many possibilities. It would also be possible to adjust the duration according to path length and to have a certain pitch for each part of speech.
The primary output format of the program is a text file that can be fed into ascii2midi. The latter then creates binary MIDI data from it. MIDI is easiest explained »sheet music for computers«. Using a sequencer this sheet music can be played by synthesizers. Furthermore, one might print out the music on a sheet, just as the term sheet music suggests. For the sentence shown in the above figure, the score looks as follows:
Sheet music created from sentence s1 of the BROWN sampler
Note that 12/8 is not the correct metre. It just ended up there for some reason, maybe it was an auto-detect failure. The red bar lines indicate that it is not possible to fit the melody into the bars if the shortest note to display is the demisemiquaver (1/32 note). The last note before that bar actually is longer than it is displayed. Displaying the hemidemisemiquavers (1/64 note) would lead to other oddities. Of course, reading the above figure is a bit confusing to people like me who have only basic skills in reading score. There is an audio example in the next section.
From Score to Music
As mentioned before, the output of the program consists of MIDI data, a kind of digital sheet music. Most musicians will agree on the fact that the score on the sheet does not form the music. There are many different ways to interpret the score. When a pianist plays a classical piece it will probably be very different from another pianist playing the same piece from the same sheet. In a certain sense, this is also true for MIDI data. It depends on the synthesizer playing the digital notes. Some synthesizers will disregard volume information entirely. Others will make the tone sound more bright but not louder. Some may have a long attack on notes so that the short notes are not heard anyways. After the note has ended some will have a long sustain mixing into the next note.
I liked to have a non-artificial sounding instrument so I picked a flute sample delivered with my Yamaha A4000 sampler. Taking the above linguistic tree and the described feature translation rules plus the named flute sample, the depicted score sounds as follows:
After several adjustments to the original rule set I achieved to obtain tunes that sounded like music. From very many themes (over 60) I picked the ones that fitted my piece best.
Final Remark
Many listeners would not have guessed that the melodies in the track Computational Linguistics were not created by a human but by a machine. Still the created melodies are unusual. There is no groove or other adjustment to the bars and a metre. Moreover, I selected the themes fitting best. There is no general solution for creating themes or melodies accepted as musical by humans for all syntactic analyses there are.
Parsing is the technology of creating syntactic analyses automatically. This way we could listen to all kinds of text as melodies, not only to those from treebanks. The question whether this will be enjoyable or not is to be answered by further experiments. Opinions on »syntactically generated music« will in any case differ. But so they do on freejazz.
Es ist soweit. Viel Blut, Schweiß und Tränen, noch mehr Gigabyte Speicherplatz, viele Megabyte Upload. Eine Menge Maschinerie und noch mehr verwürfelte Gedanken, wie man aus dieser das herauskitzeln könnte, das dann dies und jenes vielleicht bedeutet, ausdrückt, oder einfach nur gut klingt. Und alles nur für Euch. Wer auch immer Ihr seid.
Auf der Homepage des Projekts DrNI: EM steht alles weitere. Dort gibt es sogar eine Vorschau, bzw. ein Verhör, erm eine Vorhörfunktion. Und natürlich das ganze Album als OGG, MP3 oder FLAC.
Natürlich sollte hier noch Track 6 gesondert erwähnt werden: Computational Linguistics beinhaltet nicht nur von Computern gesprochenen Text sondern auch eine Melodielinie, die aus linguistischen Analysen von Sätzen aus der Penn Baumbank erstellt wurde. Unabhängig davon finden sich diverse nette Samples, viele davon auch selbst aufgenommen und einige nette Synthesizer. Wer die autobiographischen Komponenten erkennt, der kennt DrNI schon ganz gut.
Viel Spaß. Und schreibt darüber. Gerne auch schlechte Kritik, das ist auch Werbung.
