Infinitive Melody Motivation

4 people from different universities, having courses in different subjects as well. Using AI neural networks, studying the melody of any type of music. Create/generate motivational sitch for musicians

Comment

Inspiration

Recently, as AI/ML field development improved significantly, too many LLMs and NLP or image processing apps existed. The focus are limited to this area significantly. Therefore, in a less focused field of art -- music has been a less developed or focused field. As this will require professionals in many field, it is hard to generate a particularly well app. Despite a few research paper did focus in sound and midi files, most of them are for developing biometric basis.

Us developing a innovation idea of using different music as food, we can suggest some interesting sitch for music development purposes and musicans. The idea of predicting the note of the next beat for the musician to aid their development.

What it does

Our app receives the notes or patterns of a music piece or an incomplete music piece, then produces a few potential sitch based on the type of music. The suggestion is produced similar to a content-based Recommandation system with two hidden layer neural network. We auto extract from musical midi files and found the potential speaches. Using a weighted feature of classical ancient scale dataset, we can obtain a better prediction for specific emotions, such as introduce bias to sadness, hapiness.

How we built it

With LSTM(long-short-term memory) artificial neural network model, to train a model and using a public bytedance newly developed dataset in classical music. This AI model consists of two hidden layer and a final node. We used cross valiadtion and ensembling methods to predict for a better accuracy. The hyperparameters were being looked at a lot, but we decided that the default is good enough and due to time issues. We also use manually created scale dataset and midi files to analyze some interesting features. We created a fabulous website which is able to communicate to the audience. Some of these are still underwork due to the timelimitations, as our team members came from all around UK and will need to go home for rests.

Challenges we ran into

  • It is a relatively developing field of music, not much references can be found on internet.
  • We didn't have any experience with midi files, LSTM artificial neural network.
  • We didn't know a lot about AI/ML models.
  • people hae to go to different cities for rest in evenings.
  • flask is hard to learn within an hour. ## Accomplishments that we're proud of
  • We successfully created models for deep learning network with some manual constaints.
  • We created a user-friendly website where users can input their melody pattern for motivation. ## What we learned
    • flask
    • web-scrawling
    • python
    • sklearn
    • visualisation method for ML analysis such as Seaborn, matplot.
    • collborating in a team as we aren't familiar with each others
  • Research online for datasets ## What's next for melody motivation
  • Make audio file friendly.
  • Include more variety of music and musical datasets.
  • automatically play the selected sitch of music.
  • allow for the download of music.

Built With

Share this project:

Updates