ABOUT THE TEAM

We are Seoul Wave Factory, a Machine and Deep Learning-based audio and music research team from Seoul, South Korea. In keeping with the word Factory, the team consists of collaborative workers to produce quality outputs based on various musical and technological backgrounds - music producer, audio engineer, data scientist, and graduate students at MARG (Music and Audio Research Group) of Seoul National University (SNU). In the contest, we tried to seek a new direction for music production with ML and DL. Specifically, we aimed to explore whether AI could be a partner for musicians creatively and efficiently.

Furthermore, we wanted to experience the actual tech-music interactive workflow by ourselves. In the end, we wanted to see how cool the results of our tech-music mixture would be.

ABOUT THE SONG

In our track Voyage To Earth, we wanted to speak about our desire to travel. At the same time, we also wanted to give a message of sympathy to our friends on Earth who are suffering like us, with bright vibe and hopeful lyrics. With electro-house music, AI vocal sings in the beautiful Korean language. We only used fully AI-generated vocals in this track.

For the instrumental section, we used AI-generated beats, chords, and melody line sources to make the instrumental section. With the track, we also made an audio-reactive music video to express ourselves visually.

ABOUT THE HUMAN-AI CO-CREATION PROCESS

Learning and Deep Learning. In addition, we wanted to experience the actual tech-music interactive workflow by ourselves. In the end, we wanted to see how cool the results of our tech-music mixture would be.

Our common consensus in the human-AI-co-creation process was that AI would be a good facilitator in the music creation process. Hence, we aimed at two possible utilities of AI - increasing efficiency and creativeness of musicians. We supposed the AI could help a musician’s initial composition process - sketch phase, and make him/her possible to consider additional musical ideas and way faster. Also, we supposed our outputs themselves could be a source of inspiration to musicians, showing something beyond their creation.

We used several neural networks - Drum-RNN, FLAGNet, and RL-Tuner for instrument MIDI generation, GPT-2 trained in Korean for lyrics text generation, and SVS (Singing Voice Synthesis) Network for a vocal waveform generation.

With all the AI-generated vocals and instrument sources, human teammates have woven them into a complete track based on our clear musical and lyrical direction. Later, we made an audio-reactive music video to express our themes and messages visually.