Final Project Composition

Here we are, at the final post of this blog. I remember how young I was four months ago, and now I'm looking at two more credits for the year. How time flies. But do not dismay, for I have quite the post for you today. For my final project, I chose to create an original composition using all the skills I've acquired over this course. However, the idea actually predates this project. When I exposed Deep Purple for blatantly plagiarizing Carlos Lyra, I became interested in copyright—not just concerning things that sounded similar, but things that sounded exactly the same. This brings us to the discussion in copyright about AI voice replication. I'm referring to the kind of replication you'd find on a Taylor Swift cover of Frank Sinatra's "My Way"—a song she has never sung, but a video that exists nonetheless. A phenomenon made possible by AI voice training.

I began looking into the legality of works using AI voices like this, and the problems I found largely extend to the greater AI sphere as well. The primary issues are that the material used to train the AI comes from other people's copyrighted works, and that since the AI voices are created by a third party, the actual voice owner doesn't give consent to have their voice replicated, nor are they compensated for it. While it is not expressly illegal to publish works using AI voices, it is not exactly legal either, existing in a sort of legal gray area that I'd think is best avoided, especially when there are much more ethical alternatives.

Vocal synthesis is nothing new. In fact, it's been around for about as long as computers have, even with those made for the express purpose of singing. One of the most notable modern synthetic voice programs is YAMAHA's 2004 software, Vocaloid, which I’ve briefly discussed on this blog before. It uses the standard piano roll layout familiar to producers, so it's easy to navigate. You can program in the notes and then type in each word or syllable (depending on the language) for the desired note. This is accomplished by pulling from an extensive library of pre-recorded sounds. For reference, in Japanese, there are 100 distinct syllables that form words. Each syllable must be recorded once for every note and octave within the voicebank’s intended range. It’s a lot. But as a result, it creates a very convincing string of words and sentences, each of which can be further enhanced by editing other aspects of the voice, such as vibrato.

To get to the point, the reason why programs like Vocaloid are created ethically as opposed to AI voices is that they are made in collaboration with the voice provider. Any recording is done specifically for the Vocaloid software, and the provider licenses the recordings to YAMAHA. Whenever a voicebank is purchased, both YAMAHA and the voice provider take a cut from that sale. As a result, people who purchase the software also gain full authorization to use the program for anything, including commercial purposes, without having to split royalties with the creators.

With that out of the way, I was naturally very intrigued by this program and how it worked with composition. So when proposals for the final project came around, I jumped at the chance to incorporate what I had found into my work. I decided to create a Vocaloid song, specifically based on the style common in the traditional Japanese side of the program (since the English Vocaloid music is much different from what I normally make). I first did some research into what makes a typical Vocaloid song, which was hard to pin down since it isn't a genre. But after listening to more songs than I care to admit, I determined that the most common characteristics of Vocaloid music are emotional and symbolic lyrics contrasted by a somewhat upbeat instrumental, with the lyrics, of course, being in Japanese. Specific tracks that stood out to me were on the more Dreamcore side of things, which is the style I specifically wanted to emulate in my song. Below is a notable example of the sound I was going for:

Going off that example and the rest of my research, I began to write my song. The first thing I did was write a chord progression. To create the nostalgic feel I was aiming for, I kept a common tone throughout each chord, which helped create the sense of a solid foundation. However, the progression itself turned out to be a bit unusual as a result. The song is in C Major (despite my attempts to do otherwise), with the chords being a second inversion C Major chord, followed by an A Diminished, and then an A Minor chord, before reversing that progression to return to C Major again. I then layered a simple melody on top to fill in some space and found the perfect pluck synth to fit the notes.

To make it even more dreamlike, I added plenty of effects like reverb, delay, and bit crush, with a bit of a low pass for good measure. Going forward, I won't explain every little detail of each sound used in the song, but I will try to explain what I've done for each section. Which, speaking of, brings us to the structure itself. It's fairly simple, just following a standard intro, chorus, verse, chorus, bridge, chorus, outro pattern. What I find unique about it is that it changes tempo right at the end of the intro, speeding up going into the chorus. I used this not only because it sounded cool and because I copied the idea from another Vocaloid song I listened to, but also because I still had lyrics left to fit into the intro, and that was the best way I could think of adding them without making the intro twice as long.

Project Overview Above

Intro Overview Below

You can see that on line 4, I added a little arp that complements the chords, before layering on top of them in line 3 the same chords with a different synth. This synth actually has a really cool sort of flange sound to it, which creates a great effect when I automated the volume for the bassline, which uses the same synth patch. The bassline follows a faster rhythm and slowly comes in right before the tempo change, where it becomes the only instrument present. Speaking of which, if you were curious about what inspired the tempo change, it was a section near the end of this song where it gets faster right at the end. I used the inverse of that, where it changes right at the beginning. (The change in the example happens at about 2:25):

As you can see, I was very intent on matching songs in this style. The transition into the chorus and the actual energy present in that section are almost a mirror image of both the examples I've shown. The chorus itself looks more complex than it is, with most of the additional tracks being layers/variations of the initial chords or the arp. The latter of which is introduced in the second half of the chorus section to maintain interest before moving into the verse. The drums used are actually the same drum sample from the first example I showed, the "Amen Break," which is a very popular sample in Breakcore and Drum and Bass music. I only made minor alterations to the rhythm before layering the sample with my own drums for more punch.

The verse is really just a simplified version of the chorus. The drums are simply the Amen sample, and the bassline is taken from the flange synth from the intro. The arp is introduced in the latter half of it for the same reason as the chorus, to maintain interest.

The second chorus, which is really the first chorus because it's the first section with the actual chorus vocals, is almost the same as the first. The only minor difference is that the initial arp variation is present throughout the whole section, and the arp introduced in the second half is actually a new melody.

The bridge section is very simple—just the main chords lowered an octave, with the arp introduced in the second half of the section. The main driving force of the bridge is actually the vocals.

The bridge vocals serve to transition directly into the chorus and actually lead into a sort of third part of the bridge section, which has similar instrumentation to that of the chorus. It's a bit hard to explain, but you will hear what I'm talking about when you listen to the piece. This part then transitions into the chorus through the use of a quick drum fill. Since the instrumentation of these sections was so similar, there would be less impact when switching between them. So, for the final chorus section, I layered yet another set of chords on top to fill in even more space.

That does it for the basic composition side of the project, but there is still the whole question of vocals. I have a sort of odd process for writing lyrics that I don't think most people have. I write the lyrics first, and then write the melody exactly in line with the syllables in each line. Basically, I wrote the lyrics, and then wrote the melody with an instrument on top of the backing track while audiating the lyrics in my head. As a result, I do treat the vocals as an instrument, which bleeds into how I mix them as well. But I think it works for the dreamscape feel of the tracks I'm replicating. On top of that, since the vocals are literally synthesized, it makes more sense to treat them that way.

I used two different voice banks in this track: one being more mellow, which I used for the verse sections, and a louder, more outspoken one for the choruses. Each portion of the vocals has its own set of effects applied, mostly just reverb and chorus, but depending on where they are present, they may have distortion or compression as well. The chorus vocals, I believe, have the most effects applied to any one sound in the project, with all the effects mentioned above, as well as delay and bit crush. Later, during the bridge section—in the third portion of that—I added heavy distortion to the vocals to make them sound like they were yelling. I also tried altering the delivery settings in the Vocaloid software itself to produce this effect, and the results were fine. I would have produced a better result if I had used a different voice bank, but I think it fits well enough.

The lyrics themselves are in Japanese by the way, and now is probably also a good time to mention that I'm a Japanese language minor. So to clarify, I'm not just pulling the lyrics out of thin air—I do have some grasp on the language. I am actually pretty proud of these lyrics, though they are a little cringey. Here are the lyrics, as well as a rough translation, below:

Japanese:

あなたはあらゆる努力をしました
寝ても覚めても頑張った
ですが　うまくいきません
誰かに認められたかったのですか
あなたが本当にしたかったことは何ですか

自分の価値は他人が決めるものだと思ってた
自分が自分を大切にするより
周りの目が大事だと思ってた
でも　そうじゃなかった

光を見たくないですか
でも　本当は見たくないんでしょ
だってもうとっくに気づいてる
私はあなただってことに

そのままでいいんです
そのままでいいんです
そのままでいいんです
そのままでいいんです
(2x)

光を見たくないですか
でも　本当は見たくないんでしょ
だってもうとっくに気づいてる
私はあなただってことに

English:

You made every effort.
You worked hard, whether you were asleep or awake.
But it didn’t go well.
Did you want to be recognized by someone?
What is it that you really wanted to do?

You thought that your worth was something determined by others.
You thought it was more important to care about what others think than to care for yourself.
You thought the opinions of others were important.
But that wasn’t the case.

Don’t you want to see the light?
But really, you don’t want to see it, do you?
Because you’ve already realized it a long time ago.
That I am you.

It's fine the way it is
It's fine the way it is
It's fine the way it is
It's fine the way it is
(2x)

Don’t you want to see the light?
But really, you don’t want to see it, do you?
Because you’ve already realized it a long time ago.
That I am you.

Without further ado, here is my composition:

This was a very fun project to take on, and overall, this was a great class. I'm glad that there is a course here that offers exactly what aligns with my interests in music. I really enjoyed having you as an instructor, Dr. O'Leary. This has further interested me in this topic and while I do not have time available next semester to take the next level of this course, it is something I will definitely look into for the future!

Recorded Sound Project

Initially, when I started this project, I had no idea what I was going to make. The only instrument I can play moderately well is piano, but I wasn't confident enough in my classical composition skills to create an interesting solo piano piece. Nonetheless, I started writing down some simple piano melodies in my DAW, and eventually, I came up with a melody that I liked. I then began building off of that until I had three more ideas that worked with the initial melody. At that point, I was ready to actually record the melody on a live piano.

This process wasn’t too difficult to start—I just had to connect the recording device to my computer and set my computer mic input to the device to begin picking up audio in FL Studio. The problem I faced was an issue with the DAW itself that I’ve come across before when recording audio, which has sort of deterred me from the process in the past. The issue was a kind of delayed and inconsistent audio input from the device. I realized the cause was a mismatch between the DAW’s sample rate and the recording device’s sample rate. After I fixed the DAW’s sample rate to match, I was able to capture audio perfectly.

Recording the parts was easy. All I had to do was place the device on top of the piano and let it record the audio—then I could go back and correct the tempo later. I hadn’t actually heard all my ideas played together until I layered and quantized the audio files, and when I finally did, it sparked the idea that led to what I ended up creating. I could hear the potential for a brass melody that would complement the top piano melody—and then I basically just ripped off Nina Simone’s Feeling Good.

You can see the audio recordings in the top three rows of the video. I use these throughout the song as the basis of the melody and progression. Of course, the recording wouldn't have worked too well as it was, and I did some processing to make it match the timbres of the digital instruments. The song's opening was the least challenging, as I only needed to minimize the little amount of splice popping and then add a vinyl filter to make the transition to digital more impactful. I added some reverb for warmth, and also as a sort of blanket fix for the popping, which I then layered with some EQ that I used to cut out the frequencies which caused the issue. Then I added the vinyl filter on top so that all the processing would be affected by it.

The piano top melody keeps this processing throughout since I thought the broken sound this created fit better with the atmosphere I was going for. However, when I reintroduce the chords and main melodies later I added some different effects to make them sound more digital. Reverb of course is always welcome in my mixes, and then I some slight distortion for a bit of a crunchy sound. Then to reinforce the distortion I added some bit-crush on top of it to make it sound really digital. 

The ending of the sound also has some effects, but they were on the master channel rather than the instruments themselves. It's kind of an odd atonal ending, but I like how it sounds, I don't think a melodic ending would have fit any better. First I automated the tempo which created the pitch bend effect for the audio recordings, and then I added some delay and more bit crush which stopped abruptly at the end. This was partially a move to extend the piece to a minute and thirty seconds, but it was also what I had in mind anyway. Overall I really like how it turned out and was surprised how well I was able to incorporate the recordings into this piece. Now that I have the process figured out I'll probably start using more recorded audio into my music.

Mixing Project

Going into this project was actually quite intimidating; I am by no means a mixing engineer as you can probably tell from my MIDI Remix. But I took this as an opportunity to practice my skills in mastering. For most of music career I produced exclusively instrumental music, and it was only recently when I started incorporating vocals into my work. And from all that I've learned, mixing vocals is hard, especially when trying to keep a natural tone on the mix and not over processing it. For the project I chose the song "Fall Down" written by Toad the Wet Sprocket:

You can clearly tell that this song is unmixed. The first things I noticed was that the guitars were way too overbearing, and the lead vocals were in the background. And as such, the background vocals were on the same level of the lead, which is not what you want. I made a perceived space graph of what I heard from this demo:

After analyzing what I needed to fix, I immediately started working on my mix of the song. My goals were to make the lead vocals consistently in front while not themselves being overbearing, keeping the guitars more or less as is, and then placing less emphasis on the background vocals. You can sort of see what I wanted it to sound like here:

And on that note, this is what my mix sounds like:

I'm very proud of how this turned out, I think the mix sounds great, and even without directly comparing the two I think you can tell. The vocals are much clearer, and background vocals serve as intended, and the guitars are right where they should be. But you may be wondering why there are gaps in between the background vocals. This is because you could hear some guitar present in the background of the audio, and while the effects I added for the vocals sounded great, they made a weird echo when on the guitar and so to minimize that effect I removed the parts without vocals.

Interestingly enough, something I noticed while working on this project was that (while still a challenge) it is significantly easier to mix acoustic instruments. Whenever, I had mixed vocals with my music before it was a challenge because the vocals were a completely different sound to the synthesizes and heavy compression used in EDM music, but the instruments in Fall Down just sound like they work together. This is probably because everything was recorded in the same room and is an audio track rather than a direct sounds from a DAW, giving them the same level of clearness which makes mixing them much easier. 

Though a challenge that did occur early on was actually the something to do with the vocals themselves. During the recording of the lead vocals, I guess the vocalist wasn't using a very good pop filter, so there are some audible puffs of air from certain words which you could hear. I tried to remedy this the best I could while keeping the tone of the vocals by removing as much low-end as I could using the OTT compressor. This worked somewhat, and they are not as blatant, but they are still there. There is only so much I could do retroactively to fix this issue, but hopefully it wasn't too noticeably until I mentioned it.

On the topic of the lead vocals, you can see below what I did to get them to sound clearer in the mix:

As I mentioned, the OTT compressor is there. It is a multiband compressor which I used to minimize the vocal popping. I also added a very very slight bit of reverb because there will never be a time when reverb doesn't sound good in a mix. A bit further down you can also see the route knob which I have outgoing from the lead vocal channel. What this does is it affects how much to vocals are pushed over the rest of the mix, and I have this connected to an automation track which made it louder or quieter depending on certain parts.

The vocals to the song are the tracks which went through the most adjusting, and the background vocals actually needed more work than the lead:

Once again I added compression to the BG vocals to make it sound cleaner, but its not as audible because of the rest of the effects added, but its always better to start with a clean sound than to have to drown out the muddiness in reverb. There was also some clipping present, which I fixed by putting a band pass on the vocals. And finally to give it the more distant sound, I lowered the volume a bit, and added a good amount of Reverb and a slight hint of delay. This is actually only for one of the voices in the background, since their were two tracks which I mixed separately due to the differing timbres, but the other track followed a very similar process.

For one final showcase, here is a direct comparison of both the mixed and unmixed versions of the songs:

Song Production Analysis

For this production analysis I chose a song that I discovered fairly recently called "Believe it or Not" by The Royston Club. They're an indie band out of the UK. The specific version of this song is taken from their debut album "Shaking Hips and Crashing Cars", note that the single version of the song is an earlier recording and thus has different mastering than the album version.

Just off of a quick listen I can tell that there is some interesting mixing going on here. The space between the instruments isn't exactly consistent. I mean that during the verses and the bridge the instruments sound closer together, but during the chorus they sound more like a real band layout would sound where you get the full stereo audio with one guitar in each ear (or maybe just one guitar with a chorus effect on it). I made a perceived space graph based off of the chorus mastering since I believe it is more accurate to how the song was actually recorded.  

As you can see, the vocalist is in front, with the guitars panned to each side with the background vocals, bass guitar, and drums in the back. However, I do know that the vocalist for the Royston Club plays the lead guitar, and so I think that the panning is actually just the rhythm guitar mixed to stereo in both ears. This is because when you heard the lead guitar in the verses, its at about the same position that the vocals are. The BG Vocals and Bass Guitar are interchangeable since the Bassist is also the background vocalist, but the drums are slightly further out of the way as to not overpower the rest of the instruments. I was at a show last night and despite being behind the singer, the drums were still very overpowering because of how close he was to the front.

For the actual processing done to the instruments, I think that it is very minimal. It is of course noticeably different from the original single version, but that could be due to the slightly different arrangement and recording quality. There is definitely a hint of reverb added to both the Lead and Background vocals, which I think is pretty standard in most vocal music, I personally always add reverb to my vocals. There might be some slight EQ used, I think the Vocals, Guitar, and Bass all sound a bit crispier than they would if heard live, so they've probably added a small boost to the high end. The drums though I think have actually be run trough a bit of a lowpass filter, not a very strong one because you can still hear the high end but I can make out more of the lower sound than high sound in the drums. Finally, it is a bit difficult to tell if there was compression added intentionally or if it is just the standard YouTube compression present on any video, but it is there. Outside of the YouTube compression, if they did apply some, it would probably have been very slight to make the instruments kind of blend together better. I'm not sure how to explain the technical effects of compression, I'm only familiar with how it sounds when I use it so I'm just applying that to this song, but you can certainly hear some of it in larger sections of the song like the chorus.

Overall, I think the mixing is done really well, and I'm actually all for the more centered and compressed sound of indie rock. I'm a big fan of impactful drums that cut through the mix, and I think compression makes that specific sound clearer. I'm not sure how to describe it exactly, but other artists like The Rare Occasions or Jimmy Eat World have a similar mix, and I think its especially present in 90s and 00s indie and pop rock. All the sounds are clearly heard here, and I love the power of the guitar especially in the chorus. The panning just makes the sound feel are lot bigger, and I'm noticing now that a lot of other rock music I like uses the same kind of sound.