jkellys / soundboy

Sample classification app for better library exploration during music production.

Soundboy

A sample classification tool for music producers. Intended to make it easier for you to find the sounds you want.

Setup

Make sure node and npm are up to date, then it's simple as

git clone [email protected]:julesyoungberg/soundboy.git
cd soundboy
npm i
npm run dev

Project Structure

Code in main and renderer corresponds to electron's main and renderer threads. Inside renderer there is a next.js website, which is the UI of the application. Inside main there is code that manages the renderer, as well as code for DB (NeDB) interactions and heavy computation. Communication between the two processes is done with IPC, a message passing system based on channels and subscriptions. The design and function of the channels can be seen in main/ipc. The renderer has an IpcService class that handles interaction with the channels. This is made global with React Context and easily accessible with renderer/hooks/useIpcService. The main analysis code lives in main/analyzer, this module spawns a worker for classification and feature extraction.

There will also need to be a directory for python data collection and training.

Todos

• Instrument Classification
  • research instrument classification with ML
  • collect large amount of samples (probably doable with a script and the freesound.org API)
  • train an ML model for instrument classification, preferably we can find a pretrained model to adjust with our own samples.
  • load the model into the analyzer worker for sample library classication
• Feature Extraction
  • extract features from each sample (maybe using meyda - already installed)
  • save features to the DB
    • Document schema should be typed properly (currently lots of Record<string, any>)
• Interface
  • need a way to select a folder for analysis
    • send folder name as a filename to the analyzer, recursively explore directories in the worker
  • need a way to view a list of samples and filter by perceptual features
    • E.g. filtering by bright may only return samples with a high enough spectral centroid
    • need to be able to drag samples from the list into a DAW or elsewhere
  • other views/visualizations?
• IPC
  • need a new method in IpcService for sending a single message and expecting a stream of messages back (for the analysis process)