📌 Link to Alia Soundcloud

Background: Same or Different, That is the Problem

The history of music is so long that it is immeasurable, and Donald Grout, the author of Western music history, said, 'What is new is always a twisted traditional thing in a new way.' Music has come to a time when it cannot create 'new things'. In the 1990s, there was a standard that plagiarism was when 8 measures were the same, but even that has been making too much music with similarity, so now it is based on the criteria of similarity of songs, commercial use of songs, creativity of original songs.

As a result, making new music was an immense burden for many music industries, and the amount of work varied greatly depending on the skill. Just like a painter drawing a picture. Without learning music, I came up with a hypothesis.

• If you imagine the future, an AI-based music label could exist and completely change the existing music industry. If our label analyzes verified sounds and utilizes generative music technology to create new music on top of them, could we produce more music than any other label? By doing so, we could secure more copyrights, pursue unprecedented creativity, and create a unique label, couldn't we?

Tools like Midjourney already allow unlimited image generation with just a few lines of text. As innovation in the image field has made various style replications possible using different algorithms, I believe similar innovation is possible in the music field as well. With this belief, we started the project, and this is the biggest motivation for our project.

Alia Music Label: We Make Different

Alia Music Label?

'Alia' means 'in a different way' in Latin. We have given Alia the meaning of 'using a different approach' and 'using various methods'. Alia has a similar pronunciation to Aria. This signifies AI and human creating beautiful songs in a synthesis of AI and Ria (Latin for song). Alia is a creative music label that aims for beautiful music in a different way by humans and AI.

Goals of Alia Music Label

The conventional music production process involves reference setting, melody composition, MIDI work, recording by musicians or singers, and sound mixing, and then the song is completed. However, so many songs have already been made from references and melodies to MIDI composition that 'coincidentally' similar songs can actually come out. To prevent such 'coincidental' thinking, our label plans to introduce generative AI technology, which will not only innovate the existing music production process but also present creative and unique music.

Step 1. Melody → MIDI using AI : Soundraw.io

There are many tools for producing images with AI, from Midjourney to DALL E, and many AI tools for music productivity are already on the market. Although our label originally planned to use AIVA, we chose Soundraw.io. The reason for not choosing AIVA and choosing Soundraw are as follows.

AIVA.io: Why No?

To someone like me, who is not knowledgeable about music, AIVA initially seemed like an innovative tool. It was possible to create high-quality music with just a few clicks, and up to 300 songs could be made for 50 euros a month. Also, it was possible to modify with internal MIDI tools. However, this tool did not meet our team's expectations.

One of the main reasons it did not meet our expectations is that AIVA did not have the value of GenAI tools. The main competitiveness of generative AI is the ability to generate diverse results. The currently famous GenAI tools generate diverse results even with the same input text, but AIVA produces too many similar sounds due to input restrictions through buttons. In addition, issues with adaptability with other programs and overall low quality led to the tool not being considered for our label.

Soundraw: Many options and low barriers

In the past, from top liners to song camps, you had to study music itself, and the process of 'composing' is not as simple as just putting random instruments together to make the music. It's a process of combining hundreds of instruments accurately to complete the first step of music production. After that, the music is put on the market for 50 to 120 dollars, and the production cost is expected to be recouped by splitting it with the singer 5:5 or 6:4. If it doesn't sell, all the time and money spent on composing becomes futile.

Express2Music: New Approach to Music

We chose a different approach. Instead of the high barriers of making music melody and MIDI, our label chose Soundraw.io, which anyone, including non-professionals, can use to create instrumental music. Instead of the high hurdles for professionals, we made instrumental music using Soundraw.io.

Step 2. AI Singers Uberduck.io

As a label with limited funds, finding a singer was the biggest challenge. For deepfaking voices, a vocal of at least 3 hours, over 200,000 steps, and over 24 hours of training, as well as an A100 high-performance graphics card, are required. The biggest problem was the issue of licenses for famous voices, meaning no matter how much training data you have, the machines for training, or the legal responsibilities were significant issues.

Indeed, the Uberduck we used made it possible to use the voices of famous singers, and it was a startup venture company that received investment from the famous US startup incubator Y Combinator. It showed good PR effects at first, but at the time we were working, we had a lot of concerns about whether to use uberduck.ai among the many tools. One of the reasons we chose Uberduck is that there were no limits to its use and it was a tool that non-music professionals could use. Even the lady next door was able to make it at a level that was enough. As long as we don't make a profit, the copyright issue of AI is a different story, and if a certain singer doesn't come to mind, the legal issue is a different story.

Step 2-1. Choosing a Beat (Your own created beat is also possible)

Step 2-2. Writing Lyrics (Possible to generate with AI)

Step 2-3. Choosing a Singer

Korean_pop_culture.mp4

...then the music is complete! The production time for this music was less than 10 minutes, and we completed the creation of copyrighted music, lyrics, vocals, and mixed sound recordings.

The recording and singing process of traditional singers incurs substantial costs and labor costs. Even if you make it at a cheap price, with hiring a lyricist, guiding musicians, renting recording studios, and mastering, it takes hundreds of thousands to millions of won and takes days, weeks, and even months. However, by using AI, we completed the entire process for just 10 dollars per month.

Step 3. Promotion and Marketing

Instead of just listening to music, can music videos be created using current AI tools? We completed the distribution by using Midjourney and Runway tools based on the completed 5 songs. Using our last album, Pray, as an example, we will explain.

👉 https://soundcloud.com/betaai/pray

Step 3-1. Creating a Main Image with Midjourney → Generating a Video with Runway

Go to the final video result

We created an image of 'robots making music' mainly through Midjourney and completed the video production based on this picture.

Step 3-2. Background Creation → Final Adjustments

👉 Go to the final video result

There are still many awkward parts in the simple Runway video, so we completed the music video by creating a filter to make it appear as a concept of music videos being displayed on a screen rather than producing it simply with Runway.

Lessons Learned from the Project

As someone passionate about music, I must honestly admit that the music I produced in the past month is insufficient compared to the standards of the market. The quality is too low to release, and I think the market will not look for such music.

Even so, I haven't lost hope for this outcome. If we consider Midjourney, which has reached the status of a unicorn company without external investment, in just one year, it developed AI that generates images that are indistinguishable from humans. Seeing such rapid development, I believe our current project will also leave its mark in history. When advanced AI technology looks back at this point in the future, I am confident that this project will decorate a page in the history of music AI. We have come to v1 of text2audio and when v5.2 comes out, I believe the music market will also be dominated by genAI technology. After all, technology is inspiring for art, and I hope this project will become part of the history of music AI.

Team Members' Reviews

• Jungbi: It was fun to challenge the picture we wanted with many experts! AI is still far away, but it is coming.

• Bongseon Hwayeonjeong: I like that I was able to produce results through the AI generation pipeline. We learned that technology inspires art.