In recent years, artificial intelligence has become increasingly present within society, with its integration into music production raising both creative and ethical questions. My paper examines the role of AI within contemporary music production, focusing on its impact on creativity, authorship, and workflow. Using primarily secondary research supported by practical primary research, this paper explores several key areas, including the use of AI by artists in their creative processes, the incorporation of AI into my own personal creative practice, and the effectiveness of AI tools designed to assist modern music production. In addition, I conduct a practical analysis on two leading AI-generated music platforms to evaluate their capabilities and limitations. My findings indicate that while AI offers significant benefits in terms of efficiency, accessibility, and creative support, it also presents challenges regarding originality, artistic identity, and ethical use. My paper concludes that AI is most effective when used as a collaborative tool, rather than a replacement for human creativity, and that its role in music production is likely to continue evolving alongside technological advancements.
AI Contrast Between Generating and Enhancing
Earlier in 2025, a band named “The Velvet Sundown” made an appearance out of nowhere, quickly gaining over 850,000 monthly listeners on Spotify and their top song “Dust on the Wind” has nearly four million streams. However, despite their success, it was recently revealed to be an entirely AI-generated project. Before all this, however, the band denied all claims that they were completely AI-generated. Listeners had claimed that there are no individual social media accounts for the band members and there is no evidence of them playing live. Although these two points can be plausibly excused, a post appeared on social media under the band’s name with an obviously generated photo of them “playing live” with guitars having four strings and no pickups, microphone stands appearing out of thin air and mixing console faders being smushed cubes. Their Twitter (X) bio even read “Yes, we are a real band and we never use AI #NeverAI” (StereoGum, 2025). This statement is very clearly incorrect and doesn’t take a full listen to one of their songs to see (or hear). The band’s spokesman Andrew Frelon revealed he “knew this would be unbearably strong rage-bait for Twitter users” which began The Velvet Sundown’s name going viral and receiving much more attention, leading to a large increase in streams and monthly listeners.
Later on, Frelon admitted that The Velvet Sundown is generated from Suno, an AI music generation software, and that the fake bot plays had somehow caught the attention of the Spotify algorithms and began to place some of these songs into user’s daily mix playlists that are created by Spotify themselves (Rolling Stone, 2025). He runs the “official” social media accounts and on July 5th posted a vital statement on Twitter with the headline “Not quite human. Not quite machine. We are somewhere in between.” (X:@tvs_music, 2025). The statement went on to admit they are “a synthetic music project guided by human creative direction, and composed, voiced, and visualized with the support of artificial intelligence” and that this project “isn’t a trick – it’s a mirror. An ongoing artistic provocation designed to challenge the boundaries of authorship, identity, and the future of music itself in the age of AI.” This statement revealed more than an admission, but a social experiment about how far AI has come in the field of music, and how the majority of listeners absorb generated music with no idea that the “artists” are actually a bunch of algorithms run by a machine.
In my opinion as an artist and producer, this experiment proved scary. It shows that without a trained ear, AI music can fly under the radar and be perceived as human music, especially in the electronic genres where usually only synthetic instruments and samples are used; but to have a “rock band” be that successful without any streaming platform or lawyers take action, or people being able to instantly recognise the music is fake, is frightening. It could put a lot of producers out of work with the accessibility and cost of these generation sites, and they’re only going to get even more indistinguishable as time moves on.
For some artists however, AI has proved useful outside of full song generations, mainly by assisting their workflow and enhancing their musical and songwriting abilities and outcomes. Holly Herndon is an American experimental musician who teamed up with her husband and fellow musician Mat Dryhurst to create “Spawn”, an AI tool which was used on her 2019 album “PROTO”. This tool aims to mimic, interpret and develop audio that is inputted using human voices in place of instruments. It isn’t like any other generative AI tool, as it takes very specific inputs rather than learning from large datasets. Herndon trained the AI on her, her partner and her friends’s voices initially, before inviting around 300 willing participants to perform takes that would act as datasets for the AI to consume (Red-Eye, 2022).
Her track “Godmother” from PROTO is a good example of Spawn, as the entire song is created with it. They fed Spawn some percussion tracks by experimental electronic musician Jlin and a result was produced using Herndon’s voice, which was then edited for the final mix of the track (Art News, 2020). After listening to this track, the result is nothing short of weird. You can tell the original source was percussion, as the sounds are very choppy and short. However, with the replacement of her voice for these percussion sounds, the song sounds like a bunch of random human noises mixed with various reverbs and delays. Some of the sounds it produced doesn’t sound like a human at all; more so a granular sample of a noise which has been pitched or filtered, in a similar way that one would use GRM tools in electroacoustic music. In it’s defence, the track does sound robotic. Not in a tone sort of way, but more in the way these voices are being outputted and processed. I can see how someone could record their own vocals and spend days editing and manipulating them to get this result, as the outputted voices do sound quite realistic, but the way individual voices interact with one another is something a human would find challenging to accomplish.
Herndon is adamant that she doesn’t want to use spawn as a composer, but rather a collaborator, not wanting this creation to be used as a virtual instrument that replicates voices. “We’re trying to view Spawn as an ensemble member… an ensemble made up of humans and machines” (MusicTech, 2019). She goes on to say “we have material that we would like to be performed or interpreted through both human performers and AI performers” which is rather intriguing. Unlike the vast majority of people who incorporate AI into their music with composition and arrangement, Herndon is thinking in an entirely new manner. She aims to incorporate Spawn as a fellow stage member; someone (or something) to assist her in her performances of pre-existing songs she wrote. In the eyes of a performer, utilising AI this way could revolutionise live work, similarly in a way computers massively changed live electronic performances. Due to machine learning algorithms, AI has the capability to adapt to any music style or genre that you throw it’s way, which could open up possibilities on stage. I can see this development significantly helping underground artists, that may have rich material to perform but no money or contacts to be able to do traditionally. There is a lot of scope for using AI for this purpose and is something I myself would be much more onboard with in terms of AI in music compared to the current situation of using AI as a composing tool rather than a performing tool.
Another artist who has used AI creatively is Youtuber Taryn Southern, who released an album titled “I AM AI” and claims to be the “world’s first AI-composed music album”. At first glance, it seems like Southern used the 2018 equivalents of tools we have today, with IBM’s Watson Beat, Amper, AIVA and Google Magenta all being used across this album. However, it is actually more in depth than what it seems. Southern directed parameters such as BPM, rhythm, key, mood and instrumentation. After generating results with these parameters, she then arranged full compositions in a similar fashion to how you would arrange samples, then wrote vocal melodies and lyrics afterwards to fit the vibe of the song. (Taryn Southern, 2024). This is quite unlike modern generators, as she still had a human input, controlling and morphing the AI results into completed tracks pieced together by a human. Ethan Carlson, a human producer, also assisted with this record, handling vocal production, mixing and mastering. To me, this is key information because it once again separates her experiment with modern generators; she still had the human input of mixing and mastering these songs, as well as using her own human voice over the AI-generated instrumentation. This is something modern generators do for you, giving you fully completed compositions that are already mixed and mastered.
In an interview, Southern discusses the use of AI within her work and says “it’s not like you just press a button and a beautiful song is created. There is a certain amount of binary decision making by the human” before the tools generate anything, then once they have “you pick the ones you like and [then] dump the ones you don’t. It’s then up to me to arrange the pieces into a song structure to fit the lyrics.” (Forbes, 2017). This method of working still seems like there is no input from the human in terms of the presented material and, in reality, there’s not. However, she goes on to admit that she doesn’t “have a traditional music background, so having the potential to create something, from scratch, by myself, that sounds good, is incredibly empowering.” which raises another interesting point. There are many people out there who would love to create music but lack the fundamental skills to do so. Southern is telling us that she uses AI as a form of expression, in a similar way a guitarist may use a guitar or a vocalist may use their voice. Furthermore, she compares AI to other humans, saying “the process of working with human collaborators is quite similar. It’s different every single time, but it can be tricky to find the right partner who understands your vision and is reliable.” (Forbes, 2017). What seems like a thoughtless and bland album from the outside is actually a lot more human than we would think. In this scenario, she is using AI how a producer would use a human: listening to their ideas, taking the good ones and turning said ideas into completed tracks using traditional production methods. In my opinion, this way of working entirely separates the workflow of using generators like Suno and Udio that dictate the entire creative process, with the only human input being the prompt. Unfortunately, there is still something to be said on the originality of the generated results due to machine learning algorithms essentially stealing and combining pre-existing works, but producing in the way Southern did can significantly hide this flaw.
Rivers Cuomo, from the band Weezer, is yet another artist that likes to use AI in his work, except he doesn’t use it like the previously mentioned artists. He uses ChatGPT to generate lyrics as an experiment to compare them to real lyrics. The first experiment of this was a version of the 2022 Weezer song “I Want A Dog”, where he released the lyrics to the song to fans to figure out which set were generated. “I asked GPT-3 to write a song by the same title to see how they compare. Can you tell which is which?” says Cuomo on social media (ABC Audio, 2022). Although both versions of the song feature lyrics about wanting a dog (surprisingly), the AI-generated lyrics are easy to spot. GPT-3 seemed to take on a more literal approach, talking about the emotional benefits of having a dog, whereas the real lyrics dived into something used more as a metaphor and spoke of personal struggles which AI could never infer by itself (Genius, 2022). Although Cuomo has used AI to generate lyrics many times, he has never used them in a publicly released song, saying the “emotional depth” of the lyrics haven’t quite satisfied him as a songwriter (UPROXX, 2024).
AI and My Creative Practice
My creative practice and production method is reminiscent of the 1960s and 70s. I love to record things fast, in one or two takes, using minimal mics and utilising bleed. Doing things this way allows the music to breathe and exist in a way it cannot if done any other way. I also love to allow mistakes into my recordings; not huge mistakes per say, but more little nuances or imperfections which, for me, is what makes a recording feel human. All those 60s and 70s records have a natural, earthy vibe to them, as if you are in the room while the band are playing the song to you. This type of sound can only be achieved with certain microphone techniques and the presence of other band members playing their instrument in the same room at the same time. Furthermore, this sound can only be achieved using the right equipment which is why I only mix my recordings using analog emulations of famous era-accurate EQs and compressors, along with a tape machine plugin on every channel. Analog equipment and tape was a huge part of the sound back then, so without it the songs produced today don’t have that same mojo and underlying energy. As I can’t afford all the real gear, I emulate it in a DAW as best I can while utilising the many positive aspects and quality of life upgrades that a computer has over a tape machine. I took on this approach thanks to an interview I read with Ryan Hewitt and John Frusciante, talking about their production method for all his 2004 solo efforts, in which Frusciante states “it’s so clear that if you want to make records that sound as good as all those records in the ’60s and ’70s, it’s not going to happen on a computer. I’ve seen people’s skills go a long way and people being able to get good vibes out of a machine that doesn’t have a lot of vibe, but it’s never as good as what that same person would do with tape.” (Tape Op 61, 2007). This leads me on to how AI integrates into my practice.
As Frusciante rightfully stated, making a 60s record on a computer won’t make it sound like the 60s, purely because nothing was digital and sounds were manipulated by capacitors and op amps rather than binary which already gave the music a more grounded element. However, plugins have come very far from when digital was first released and almost every company has released a product that aims to give that analog “warmth and glue” that everyone is now missing. Companies like Universal Audio and IK Multimedia are renowned for doing a phenomenal job at replicating some of this vintage and rare gear. One product that is doing things a little different however is TAIP by Baby Audio, which labels itself as an “AI-Powered Tape Saturator” and states that “unless you want to maintain a 600-pound hardware device in your studio, TAIP brings you the closest you can get to the reel thing” (Baby Audio TAIP, 2021). This is a pretty bold claim, but the use of AI here proves to be interesting. Baby Audio say AI is used here in place of traditional DSP which is commonly found in plugins. This stands for Digital Signal Processing, which unlike it’s counterpart ASP (Analog Signal Processing), “uses algorithms to process and modify digitally converted signals (discrete values)” compared to using “physical components like resistors, capacitors, and amplifiers to modify signals in real-time (as continuous values)” (LEWITT, 2022). DSP “would entail guesstimating the effect of various analog components and their mutual dependencies” and therefore have trained an AI algorithm “to accurately decipher the sonic characteristics that make a tape machine sound and behave the way it does“, (Baby Audio TAIP, 2021). Casper Bock (Baby Audio’s boss) has stated in an interview that “to get a computer to behave (or sound!) in a certain way, it helps to think like it does. Re-creating an ‘analog-style’ signal path in DSP is thinking about the problem like a human. The AI approach helps us solve the problem like a machine would” (Attack Magazine, 2021) which was Frusciante’s problem I mentioned earlier. This use of AI helps the computer to think in an analog way, rather than it’s standard digital way which I find extremely interesting.
As for music generators, they are extremely far off at the moment. The AI struggles at the best of times to adhere to every command you put in the prompt box and will often take your demands as a suggestion and end up doing it’s own spin on your idea (I talk more about this a later section). AI heavily struggles with details and simply looks at the broader areas in your prompt. These records were made with very specific microphone techniques and placements, microphones themselves and equipment, but most importantly: the musicians. This one factor alone is why I am not threatened by AI in terms of the music I want to create. No matter how hard AI will try (right now), it won’t be able to capture the essence of a group of real musicians gathered in a room together and playing off and reacting to each other. Instead, it will shove multiple characterless MIDI instruments together and play perfectly on time as if a band of robots were in the studio. If I ask it to create a 60s rock song with specific details such as Glyn John’s drum technique with U87 microphones entirely mixed through a Neve console and to half-inch tape, it will simply just create a 60s rock song and disregard all my other commands. This is where us as producers simply have the upper hand. We can record and mix a song in whatever way we choose and craft our signature sound, whereas AI already has it’s signature sound: completely digitally clean.
AI has come a long way, even from five years ago where it was significantly inferior to what it is today. But in my opinion, an algorithm will never be able to capture the vibe of real musicians playing real instruments and real producers using their desired equipment to finish a song or record. I do believe, however, that AI can be utilised to assist in getting that analog sound like TAIP is currently doing. With fine-tuning, an algorithm could recreate the nuances and quirks of analog gear and tape with precision accuracy, but we will have to wait and see how it evolves in this field over the coming years.
LANDR
LANDR is an “all-in-one” digital music platform, offering VSTs and plugins, sample libraries, streaming service distribution and AI-powered mastering. The mastering service comes in the form of a plugin or an online website and uses an AI model trained by professional mastering engineers; it’s used by the likes of Lady Gaga and Snoop Dogg. The service “uses AI to pull knowledge from thousands of mastered songs to offer a unique mastering chain for every track” (LANDR, 2014). The plugin version of LANDR has many more controls, perfect for fine-tuning the sound, but the online version is very simple and doesn’t have many options, largely focusing on the two “style” and “loudness” ones. “Style” has three options: warm, balanced and open. Warm replicates vintage processing with softer compression, balanced is controlled with the focus on clarity and depth, and open is modern-sounding with an emphasis on punch and presence. “Loudness” also has three options, low, medium and high, which alter the balance between loudness and transient integrity respectively. The tool also features a “reference” tab, which allows you to upload a song you want your master to sound like. The AI will analyse it and adjust parameters in your track to get it as close to the reference as possible. Not only can it master a song, but can master an album of songs to keep consistency across your project and optimises them for streaming services, producing mp3, wav and HD-wav file types.
After reading user reviews, surprisingly hardly any were positive. Users describe the service a “ridiculous concept”, as mastering is not a “one-trick pony” and heavily benefits those who can mix at a professional level. This is because LANDR seems to use a very copy-paste formula when it comes to mastering, not taking your own track’s sonic characteristics in mind and just applying certain processing every time. Although it gives “useable results”, many reviews encourage you to use a human mastering service or learn the basics of mastering yourself as it doesn’t do a very good job at listening to what your track needs. Without personally using it, this seems like a tool useful for either professional-level mixing engineers that want to get a track finished with as little hassle possible, or for inexperienced producers who have no knowledge on the subject yet want to get a “release-ready” version of their track. This is backed up by the summary of Sound on Sound’s review, saying “if you can’t afford professional mastering and lack the ability, time or inclination to master everything yourself, the LANDR Mastering Plugin should be a very attractive proposition!” (Sound on Sound, 2023). I believe AI will continue to struggle to find out what your track really needs because it doesn’t actually know what genre your music is in, it just takes an educated guess. This is why human mastering engineers will continue to reign, as they have an experienced and more reliable database of how your track should sound thanks to the human brain.
Practical Analysis
Out of curiosity, I decided to put two AI music generators to the test: Suno and Udio. I did this experiment to see how authentic and capable these generators are and to see how accurate the music generated was from my prompts. I tested multiple prompts on each with varying styles and requests.
First I tested Suno and asked it to create “an irish folk song with a dubstep bassline, 140 BPM, upbeat”. It came back with four results, two of which are labelled “v5” which seems to be the more advanced version of the generator, but was limited to only a minute preview behind a paywall, compared to the other older versions which were full tracks. For this analysis, I will only be talking about the v5 previews as they were the higher quality of the four. When I listened to the result, I was honestly surprised at how believable and somewhat good it actually was. The acoustic irish folk instruments were recorded well and I couldn’t hear any artifacts or aliasing. When it switched to being dubstep, it did it very well: the original riff repeated itself but became almost synth-like, the drums and bass were sidechained creating a pumping effect, the overall sound was very balanced and mixed well and was upbeat 140 BPM like I asked; I can’t find a negative about this. With the next prompt however came problems.
I asked it originally to “combine Aphex Twin with Fugazi” and it came back with an error, saying that I can’t use specific artist’s names. This was interesting to me, as it clearly can take inspiration from these artists but when you mention them by name, they refuse to generate music they would’ve generated if you said “combine acid house with Washington D.C. punk from the 1990s”, which is exactly what I did next. For what seemed like a simple prompt stumped the generator, as it gave me solely acid house music. The generations were believable and clean like my last prompt, with the filter even being moved around on the 303 rather than being a static sound, but it had no 90s punk influence at all. I will also point out that the two generations from the older model had even more problems. At times, there were random robotic drop-outs and weren’t even in the style of acid house; just typical house tracks which was even more disappointing. I decided to get a lot more specific with my next prompt and asked it to make “a roland 303 bassline with distorted punk guitars, with a male vocalist singing how much he loves AI”. What did I get? A generic, Green Day-esque instrumental punk track. No 303, no vocals. If I asked for that sort of track, I would’ve been somewhat impressed; the guitars sounded open with the rhythms and leads being separated well and the drums sounded multi-mic’d and compressed. The biggest letdown was the feel of the drums: it was extremely robotic and had no variation with velocity, as well as the fills sounding very MIDI-ish and fast, meaning it would be impossible to play with that accuracy.
Since it couldn’t generate lyrics in a prompt, I used ChatGPT to write me some lyrics about AI taking over the world, then copied them into the lyrics section of the generator. I asked it to make “a dark, cinematic acoustic ballad, 80 BPM, downbeat, minor key”. Both results gave me practically identical structures and instrumentation, just with alternative chord progressions. This prompt tested two main questions: the quality of acoustic instrument replication and how believable the generation of the human voice is. Both pieces begin with a single acoustic guitar accompanied by a solo violin line. This allowed me to hear how well Suno generated individual acoustic instruments without being covered up by other arrangements. The guitar sounded decently recorded, but was trying it’s best to cover up the artifacts with delay and reverb, which also had significant artifacts in. The violin, which is supposed to be very dynamic, felt entirely flat. It had room reverb applied to it and other than slightly fluctuating in volume, had no human touch to it at all. I can easily see how these could both sound so much better with a lot of other instrumentation around them, but when isolated you can clearly hear the aliasing and MIDI-like qualities. In terms of the vocals, they weren’t too bad. They didn’t sound that robotic and the changes of pitch were smooth and felt almost natural. What didn’t feel natural was the vibrato; when it was present on a couple words, it felt like a switch in a sense where it just turned vibrato on and off, rather than a gradual build like in the human voice. It also was too perfect, with the pitch fluctuating in a sort of sine wave-like behaviour. On top of this, there were subtle but noticeable glitches every now and then, similarly to what you’d hear in extreme autotune, and what sounded like digital clipping present in the more expressive sections. The overall mix of this song felt off too, with the vocals being way too high and too airy while the guitar sounded clangy and muffled in most sections.
My final Suno prompt was “a death metal intro that transitions into a circus waltz verse and ends with a k-pop chorus”. The results I got weren’t quite my suggestion, as they both began with a circus waltz then went almost straight into a death metal verse and chorus section with no elements of k-pop in there whatsoever. Instead of the AI listening to me, it took my prompt as a suggestion and did it’s own thing. I believe this is an example of incompetence, knowing full well it can’t generate my idea exactly how I asked it to, so tried to satisfy me with a loosely similar generation in hopes that I’d like what it gave me and to make me forget about what I originally asked it for. The generations were yet again almost believable like the dubstep irish folk song, with synths being used in place of an organ and the production feeling polished with the double heavy metal guitars and drums. The drums however were even more unrealistic and robotic than the previous examples. There were often fast fills, which were way too fast for a human to play perfectly in time and were the exact same velocity. Both results also seemed to heavily rely on synth arpeggios and had the same effect as the acoustic ballad, where they were two different generations but with copy-pasted structures and instruments.
After Suno, I did a couple tests on Udio to see how it compared. It did far worse than Suno did, with many artifacts and overall bad quality. Unfortunately, due to a deal with Universal Music Group, they have removed the download feature, meaning I can’t reference them here. I first asked it to generate a “progressive rock song with synths, that is entirely played in reverse”. To no-ones surprise, it wasn’t reversed. Instead, it generated lyrics and a male voice that sung about reversing time. It did do a decent job at sounding like something Yes would write, especially with the harmonies, but the drums felt random and clanky. The other generation sounded like a Depeche Mode song mixed with What Is Love? by Haddaway. The one element that stood out most in both generations were the vocals. In one song, they sounded compressed to oblivion and were sidechained to the entire mix. The other song had very muffled and artifact-y vocals, sounding as if the vocals were isolated digitally using an mp3 file. Another thing I noticed was the unrealistic vibrato, where it came in on random phrases and was extremely electronic, sounding more like a wobble than something expressive.
My next idea was to ask it to generate “acid house in 7/8 time signature, upbeat, 125 BPM”. It completely ignored my 7/8 suggestion and instead created it’s own time signature. The first generation was not acid house at all, more so dubstep, and sounded nothing like I asked it. The second result did have a 303-esque bassline and featured very random drum patterns. There was also this white noise/loose snare sound constantly present in the background, and in some sections the drums or bass dropped out randomly, sounding like they had been dunked in water for a second. Both results were bad quality sounds, with a lot of the instruments sounding way too airy for their role in a mix.
Lastly, I asked it to write a “chiptune song with breakbeat drums”. It made a very tinny sounding chiptune song with a Mario Kart vibe to it and featured no breakbeat drums. Where Suno could generate good sounding electronic music with ease, Udio struggled to even make it sound as if it hadn’t been converted to an mp3 ten times.
After testing both these generators, it seems that if you use a prompt that revolves around electronic music, or even modern produced pop/metal songs, the result won’t be half bad and may actually fool the average listener into thinking its a human-made track. However, if you are wanting to generate more classic “recorded” material, this is where the generators are lacking. Artifacts and aliasing can be heard, along with robotic MIDI-like progressions with little variation without getting extremely specific. Suno did a far superior job at every single prompt I threw at it compared to Udio, but both often ignored my requests, or part of them, and proceeded to generate a track loosely based on one of the idea I had mentioned.
Bibliography
Admin, E. (2021) Baby Audio has released TAIP, an AI-Powered Tape Emulation plugin, Attack Magazine. Available at: https://www.attackmagazine.com/news/baby-audio-has-released-taip-an-ai-powered-tape-emulation-plugin/ (Accessed: 18 December 2025).
Audio, B. (2021) Taip by Baby Audio – Ai Tape Saturator plugin (VST/au/AAX), TAIP by Baby Audio – AI Tape Saturator Plugin (VST/AU/AAX). Available at: https://babyaud.io/taip-plugin?srsltid=AfmBOoq0GsYAi0_tMXcZ4C513L04LNRbCcxcp5XunInuuraTogtVYbTW (Accessed: 18 December 2025).
Baccigaluppi, J. (2007) ‘42 / Mr. Frusciante’, in Tape Op 61. Los Angeles, California: Tape Op, pp. 42–48.
Betts, W. (2019) Holly Herndon is forging ways to emote and sing with an ai, MusicTech. Available at: https://musictech.com/features/interviews/holly-herndon-proto/ (Accessed: 23 December 2025).
Chelosky, D. (2025) Ai-generated psych-rock band The Velvet Sundown Rack up hundreds of thousands of Spotify streams, Homepage. Available at: https://stereogum.com/2313501/ai-generated-psych-rock-band-the-velvet-sundown-rack-up-hundreds-of-thousands-of-spotify-streams/news (Accessed: 21 December 2025).
Cuomo, R. (2022) Weezer – I want a dog lyrics | genius lyrics, Genius. Available at: https://genius.com/Weezer-i-want-a-dog-lyrics (Accessed: 21 December 2025).
Droke, C. (2024) Interview: On ‘SZNZ,’ weezer thrives by Looking ahead, UPROXX. Available at: https://uproxx.com/indie/weezer-interview-rivers-cuomo-sznz-winter/?utm_source=chatgpt.com (Accessed: 21 December 2025).
Farrier, D. (2025) Rolling Stone gets hoaxed about a hoax, Webworm with David Farrier. Available at: https://www.webworm.co/rollingstonehoax/ (Accessed: 21 December 2025).
Frelon, A. (2025) The Velvet Sundown ‘We are somewhere in between’., X (formerly Twitter). Available at: https://x.com/tvs_music/status/1941474152235348443 (Accessed: 21 December 2025).
Cappalleti, G.M. (2022) Holly Herndon, Mat Dryhurst and their AI baby named Spawn, RED-EYE. Available at: https://red-eye.world/c/holly-herndon-mat-dryhurst-and-their-ai-baby-named-spawn (Accessed: 23 December 2025).
Houghton, M. (2023) LANDR Mastering Plugin, Landr Mastering plugin. Available at: https://www.soundonsound.com/reviews/landr-mastering-plugin (Accessed: 22 December 2025).
Johnson, J. (2022) Rivers Cuomo teases new Weezer single with ai-generated lyrics, Rivers Cuomo teases new Weezer single with AI-generated lyrics | ABC Audio Digital Syndication. Available at: https://digital.abcaudio.com/news/rivers-cuomo-teases-new-weezer-single-ai-generated-lyrics?utm_source=chatgpt.com (Accessed: 21 December 2025).
McDermott, E. (2020) Holly Herndon on her Ai Baby, reanimating Tupac, and extracting voices, ARTnews.com. Available at: https://www.artnews.com/art-in-america/interviews/holly-herndon-emily-mcdermott-spawn-ai-1202674301/ (Accessed: 23 December 2025).
Pilon, P. (2014) Ai Mastering: Online audio mastering – try free, LANDR. Available at: https://www.landr.com/online-audio-mastering (Accessed: 22 December 2025).
Schawbel, D. (2017) Taryn Southern: How this YouTube star used AI for her new album, Forbes. Available at: https://www.forbes.com/sites/danschawbel/2017/09/26/taryn-southern-how-this-youtube-star-used-ai-for-her-new-album/ (Accessed: 22 December 2025).
Southern, T. (2024) I AM AI, Taryn Southern. Available at: https://tarynsouthern.com/album/ (Accessed: 22 December 2025).
work, L.C.T. at, Tobias, T.K. and Team, L.C. (2022) What is DSP?, LEWITT. Available at: https://www.lewitt-audio.com/blog/dsp-what-it-and-how-to-use-it (Accessed: 18 December 2025).
