Music_Gen_Streamlit

"Music Generation using Neural Networks" Streamlit App

TO DO:

• Make a run_app.sh

• Introduction [~5 min] (Sohaib)

  • Team Member names/intro (WS 2019/2020, course name)
  • Outline
    • Introduction
    • Data
    • Phase 1
    • Phase 2
    • Neural DJ
  • Literature review
    • Refer to the literature slides

• Data Analysis [~10 min] (Shivaani)

  • General Info about data
    • Add charts and sources about data (Lakh and RedditPop)
  • DataProcessing Pipeline Graph (streamlit graph docs/ Abdallah's choice)
  • Raw MIDI data
    • Music21 intro and applications
    • show raw midi (without explanation, button to visualize raw midi (Sohaib))
  • Tokenized MIDI
    • Show Tokenized with variable length
    • Expand Tokens (musicautobot, go through section and see if its too explained)
  • Play MIDI (Sohaib)
    • Phase 1
      • Play original
      • Play extracted
    • Phase 2
      • Play sample (Lakh)

• Add Model() (Abdallah)

  • Phase 1
    • Modeling (Hugging Face OpenAI GPT2)
    • Data
      • Only Piano (extracction process)
    • Tokenization problem
  • Phase 2
    • Modeling (Architecture, Transformer XL)
    • Data
      • Used piano, but then ran into problems
      • handle big files problem
    • musicautobot API for data

• Add Prediction() (Code: Sohaib)

  • Overview of pred process
  • Play
    • Play original
    • Play predicted
  • Visualize
    • Note sheet original
    • Note sheet predicted
  • [?] Metrics

• Technical Service/Deployment Pipeline (Code: Sohaib)

  • make a requirements.txt
  • Docker and heroku deployment
    • Make container
    • Check functionalities for each functional part of interface

Use Docker to run

make sure you have docker installed ./run_app.sh (this will take around 7-10 mins) then go to : localost:8501

RUN DEMO

https://neuralpiano.herokuapp.com/