This project covers Machine Learning and Artificial Intelligence across many varying softwares and programs for example Runway ML and Processing. I cover consciousness and awareness of computers.
Consciousness, general attention, awareness. What makes something conscious? The mind/body problem. Qualia
I want to train a runway model based on album covers of female artists for this project. I want to explore the concept of biased training of models and by creating this project I aim to draw attention to it as well as female empowerment, equality and feminism.
During my research I found many articles about a music steaming service which removed female artists from the covers of their own albums by blurring and editing the covers while leaving males as they were on the originals. This makes me curious as to how well the model would work and if the limited data set would portray this idea in an effective way.
I do realise it is risqué to go forward with this as there won’t be many constants, most pop albums created by woman usually have a photo of themselves on the cover however some show the face some the whole person and each is usually taken from different angles meaning it is very unlikely that the model will be trained well enough to show clear faces. I do believe this could be an interesting area to explore.
The inspiration also comes from my own collection of vinyls and my interest in album covers.
I’m currently gathering my data set manually downloading images of female artist’s album covers. I’m working my way through a list of artists provided by google as ‘popular on web’ as I think this is the best way to narrow down the most relatable and known artists, it also provides a surprisingly varied result of genres and backgrounds of artists. It ranges from pop and country to electronica and soul. I decided to manually download the albums rather than using a web crawler because this way I’m double checking all the album covers are legitimate since a lot of fan art is published with edited covers. I know this extends the amount of time I will be spending on this project before I can even start training my model but I hope it’s worth it.
I’m considering what options I have after I train the model and I’ve been considering some possible experiments with available models in runway.
I will most likely scan my trained model through different existing runway models like for example seeing if the model accurately generated faces it can recognise since many female artist’s album covers clearly show their faces or maybe capturing some images from the trained model and seeing what kind of text the computer generates as captions. It could turn into a playful tool where it’s possible to produce an AI generated album cover with a title. I like the idea of the possibility of creating music through AI which is where I hope I can lead this project towards, I’m not exactly sure of the logistics behind this or how to even do it but it could be an interesting area to research and explore for me to really push this project.
Agnieszka Chylińska, Billie Holiday, Nina Simone, Iggy Azalea, Cardi B, Avril Lavigne, Barbra Streisand, France Gall, Janelle Monáe, Amy Macdonald, Lily Allen, Nelly Furtado, Kesha, Edyta Gorniak, Doda, Tove Lo, Meghan Trainor, Carly Rae Jepsen, Alessia Cara, Sabrina Carpenter, Lara Larson, Florence and the Machine, Cher Lloyd
Some sample images taken during training of the model.
I am very happy with the final trained model I will insert some screenshots from the sample video which I really like as AI generated album covers.
An entire album made with AI:
Amper music – Open source AI music generator
Running my trained model to view more varying results.
In search of titles for the album I ran the covers through a caption analyser, I then choose words and phrases from each one to create less literal titles.
I’ve been working in processing and Audition to create lil 30-40second sound samples for the albums. I sorted with processing code I found which creates sound based on an image and the brightness and darkness of a selected area. I then ran this through Audition and and played around with different effects for each album. I also recorded a translate program speaking some of the words from the titles to incorporate in the ‘music’. I used artificial female voices for this.
After finishing 3 Album samples I went back to creating the website.
The final website currently looks like this. I enjoy that it’s quite simple and the Album artwork generated by my Album cover is prominent. I also added an ‘about’ page on the website to explain what the site is inspired by and how it was created. This is obviously a prototype in it’s early stages but I’m hoping to keep creating and adding to it in the summer.
Link to the website: https://martazofiapalacz19.wixsite.com/femarti
Evaluation and Moving Forward
Overall, I really enjoyed this project. I tried to experiment with different softwares like RunwayML, Processing and Audition. I’m really happy with the concept I chose to explore during this project and creating the music, collecting the album artwork data set, discovering the generated results from my model and creating a website was actually really fun. I finally felt as though everything I was doing was enjoyable, interesting and fun. I will definitely be building on this project and continuing to edit and add to my new AI steaming website. There are still many tools I would like to play with and discover , like Amper music if I get a license, as well as creating longer pieces of music.