# Final essay, group 16

# Introduction

Our overarching inspiration is that of the dream. We would like to describe in music the multiple steps of a dream, namely the descent to the dream world, its random-like structure, and the rise from sleep back to the real world.
We choose to grasp this idea by dividing the piece into three main parts.

First, the introduction will feel like a transition from the real to the imaginary world, a descent, fall. The auditor should feel as falling into an indistinguishable mass, slowly preparing for the rest of the unpredictable story that is yet to happen. To achieve this, the piece with high velocity notes to represent the sudden fall, combined with a subtle *decelerando*, *i.e.* the pace of the music will gradually decrease, to show the habituation of this change of state. Furthermore, the pitches will go from high to low, in order to, again, picture a descent. The last decision is justified by the fact that higher pitches sound like they are spatially high. The notes will probably use non-diatonic scales, *e.g.* a whole-tone scale, and the arrangement will remain quite simple limiting itself to only a few instruments, to catch the lightness of the beginning of sleep.
Then comes the dream itself. As anyone has already experienced, this part can seem random, although it never really is. We have decided to create different parts/sub-sections that hold potentially very different emotions; thus, we do not impose an outcome onto the dream with respect to feelings. To combine the different parts, we will interleave them with decreasing durations. The listener should feel uncertain about what will happen in the next few seconds, and this will be achieved mostly through rhythm, because too brutal changes in tone might sound too disorienting/incoherent. During this part, we will also insist on panning and reverb to trouble the listener's focus and make sounds "bigger", respectively.

In the end, the listeners should emerge back to the real world. This, in contrast to the introduction part, should feel like uprising; more precisely, pitches will go upwards, and the velocity of notes and speed will climb up to achieve this effect of abruptly leaving the dream. Indeed, this is a choice, since dreams can ease out, but we would like the listener to perceive a shocking awakening.

The effect of the introduction is inspired by Saint-Saens' *[Aquarium](https://www.youtube.com/watch?v=YVpl-RNzdE4)* (near 30 seconds), where the brilliance in the descent is its actually steadiness in terms of pitch. We also esteem [Paper Mario's](https://www.youtube.com/watch?v=i13K77BAL50) (near 7:20 for example) sometimes unsettling feelings, but mostly the effort put in to adapting music to the landscape/environment; the aforementionned part has a remarkable time signature change. Finally, we will mostly use piano rolls for visualisations, because of their convenience when there are many different tracks; but pre-generated short musical phrases could be displayed with a score.

# Methods

## Introduction

The introduction is composed manually to have more control over the outcome. We started by playing it with high velocity and pitches and playing a subtle *decrescendo* towards lower pitches; we have improvised this part entirely on the piano. This achieves a very vivid effect, because the piano mostly arpegiates through chords that are usually unrelated; it thus sounds very chaotic and *literally* like a fall, as we descend to lower pitches and intensities. To end of the piano part, a flute enters and plays a whole-tone scale, starting on a half-tone lower each time, adding intense stress. Interestingly, we didn't even play this scale carefully over chords that wouldn't sound conflicting, and yet it still sounds natural and it matches the idea of chaos. Then, the strings come in, where cellos play again improvised chords, preparing us for a more relaxed ambiance. We could also see this as the fall's end, and the rise of *another* world of experience.

## Euphoria

We generate this part using random graphs: we create *n* levels of *k* vertices, and every vertex is connected to all vertices of the next level. Each node holds a scale that is either randomly chosen or which has its root in the very first scale of the tree. Then, we encode a notion of distance between two scales, where we basically penalize commonality with an arbitrary distance function. Finally, we run a shortest path algorithm to generate the progression.


*Above, you can see the instance we kept, with the corresponding chord progression.*

What has appealed to us in this procedure is that the progressions tended to be mostly regular, with a few insertions of absolutely unrelated chords. We thus enforced the presence of standard chord progressions, namely I - v in psychedelic music, while still adding unexpected elements fitting the unpredictability of dreams.

In our instance, this is exactly the case: progressions are mostly composed of I - v, but chords such as C#Maj, AbMaj and BMaj appear from nowhere.

We have emphasized these changes with choir sounds of the Melotron and a break in percussions. It added a floating feeling to it, especially on C#Maj a heavenly atmosphere emerges. The AbMaj - BMaj progression had a blurry feeling, as if the first subsection drifts away to return to a simpler feeling, the second subsection. There, we insisted on playing simpler parts, focusing on rhythm, especially cuts: they made us feel as if we had to catch our breath, for a second.

Overall, there is a lot of reverb on most instruments, making them sound floating, as if they had a lot of space, extending their sustain. It made envelopes less abrupt, producing a feeling of flowing in an euphoric state. We have extensively used reverses too: it made the sounds much less understandable, and it was also a nod to bands such as The Beatles that have used this technique for the same purpose. It worked well on the guitar parts, because, the enveloppe being reversed, the sound's ending was sharp and felt like it grew up towards the listener. But it worked best on the voices. Adding voices was already interesting as it alluded directly to consciousness, language being deeply rooted in our minds. Furthermore, adding the reverse created this *strong* sensation of being unable to grasp what happens during dreams and to verbally qualify the experience.

## Melancholy

For this part, we used a chord basis from two jazz standards (*Autumn Leaves* and *Nardis*). We split up the grids into distinct parts (known as A's and B's), which were then interleaved using a Markov chain. The transition probabilities were hand-picked to try and create a seamless overall structure. For the melody, we used two approaches: hand-written lines and a randomization of small note patterns (SNPs) in the natural minor scale, both interpreted by a piano with a "retro" processing. The melody's rythmic elements were left to the interpretation to allow for more flexibility.


*Above you can see the chosen instance of the algorithm we chose to record.*

In our instance, "la citation" refers to a citation of Paul Desmond's solo on Chet Baker's version, and *Nardis*' phrase was the original theme.

<br>
Furthermore, we wanted to focus on having each tune bring something different: *Autumn leaves*', by its tonal construction, felt more inclined to a standard melancholic vibe, whereas *Nardis*, a tune rather driven to modal improvisation, was conceived to bring a smoother, deeper feeling, rather than the base approach to melancholy. We heavily considered using more Nardis parts, but felt the tune's richness in particular was counter-productive to the overall perception of the piece when overused.

Having *Nardis* as the second part played felt both natural and surprising, as it brought a break to the "explicit" sadness of *Autumn Leaves* which played into the dream's incoherent feel, and the progression from *Autumn Leaves*' part A to B really brought the perfect conclusion to the piece. The tension was well managed throughout the piece, and was further reinforced by the phrases and outputted SNPs sequence. However, to avoid a too brutal change to *Nardis*, we applied a low-pass filter on the piano, so that the sharpness of the blue notes wouldn't stand out too strongly, but rather contribute only by their color to the piece. Finally, during *Autumn Leaves*', we added strings to emphasize the melancholic aspect.

For the melody, we liked this output as Desmond's first solo lines were fitting to this part's goal, and sounded coherent with the SNPs algorithm's output. From the beginning, we knew *Nardis*' phrase was going to stand out, but we are glad it output the theme as it was the most fitting for the grid. The outputted SNPs pleased us quite a lot as well because of how reoccuring the 6ths and 4ths were and their associated tension, always to fall back to the triad.

## Structure

Finally, to structure the entire piece, we originally thought of simply chaining emotions randomly. But this would have implied making shorter parts and having more different emotions; otherwise, we would lose the audience's attention. Yet, making more parts was too ambitious and we wanted to be able to express ourselves more within the parts themselves. This is why we came to the following idea: interleave with decreasingly shorter intervals of time the two parts we have generated. It was a big risk --- and we will discuss the results in the last part ---, because it was hard to project ourselves this far, and we didn't know what to expect. However, the idea in itself was interesting as it would make the dream feel much more chaotic, sudden, and thus make the audience experience an actually intense dream, not knowing where to put themselves in this constant input of emotions. We will discuss in the next part how it turned out and what we did with this.


*Above is the output of the interleaving algorithm. (A: euphoria, B: melancholy.)*

# Conclusion and results

[Link to the final piece.](https://youtu.be/-GLHa4SjGwg) (The piece starts at 2:04.)

First of all, we generated and composed the two emotions separately, and we believe that these two parts are decent pieces on their own. Euphoria expressed really well the feeling of being happy to the point of being disoriented; the chord progression sounded very interesting, and the "brutal" changes turned out to sound quite natural, thanks to the arrangement. The melancholy part was in some sense a bit less directly related to dreams on its own. Yet, we think that it did reflect extreme sadness with still some disorientation, especially during the middle part (when *Nardis*' chord progression was playing). Indeed, we have found that the change of progression from *Autumn Leaves* to *Nardis* and back didn't sound too shocking, but still added a great change in color and emotion to the piece --- the *Nardis* part sounding less explicitly sad.

Then, when we first assembled the pieces using the algorithm for the structure, we were first very dubious about the result which sounded actually quite artificial, and that was the whole risk as mentioned earlier. Yet, we didn't give up, because there were some great transitions and we wanted to make this work, so we started fixing the piece. (Musical debugging?) First, we added this recurrent sample that made transitions from one emotion to another; in our estimation, it already fixed a lot of issues and blended some parts together, and the sample itself was pretty interesting as it sounded like a stealth shadow, that rapidly moved, calling the listener to a rather dark part of the dream. However, it still wasn't enough, so we added two types of transitions:
1. low-pass filters, that added a very interesting effect of diving into a deeper level of consciousness or of the mind, although they sometimes could sound a bit too synthetic especially when strings played;
2. fade-ins, that worked very well, as they are simple and smooth transitions that allow easy blends of two parts; they maybe had less *meaning* by themselves, but they probably sound more natural than the low-pass filtering.

In the end, the reverse introduction sounded a bit too long and the strings part didn't take advantage of the reverse (since their envelope is pretty much symmetric), so we actually only took the reverse of the piano part, which had much more impact and sounded as if the listener was still trying to process what just happened to him.

When listening, the audience seemed to have enjoyed the piece and the experience (thing for which we are grateful!), and a general comment was that, even without saying that this was about dreams, they could have figured it out by themselves, which is maybe the greatest success of the piece.

To conclude, we are actually quite content with the result of our piece. It was quite a lot of work, but it turned out to sound pretty good.

Thank you for reading this! We hope you have enjoyed the experience as well.

--- Oscar Davis, Alexandre Doukhan, Bastien Faivre as Group 16