ReverbML

An Electron app to play, visualize, and annotate your audio files for machine learning

Operating System	Status
MacOS
Linux (Debian)
Windows	Coming Soon

Quickstart

Running Development

yarn install
yarn run start

Building

Make a compiled executable for your current local platform (will generate; ie .app for macOS). Output will be to out directory.

yarn install
yarn run package

Same as package except packages it in a distributable manner for your current local platform; ie .zip for macOS or dep and rpm for Linux. Output will be to out directory.

yarn install
yarn run make

Instructions

• Local database saved to your ~/reverb.sqlite3.
• Audio samples are saved to ~/reverb-export.

Developer Notes

• As of "typeorm": "^0.2.7", the docs are wrong for lazy loading relations. Even when property type is Promise<T> as the docs say, attempting to save the record will fail. (even though their examples says you should be able to do {key: Promise.resolve(value)}) typeorm/typeorm#2276
• Consider migrating Classification back into the Label table. Although it breaks BCNF, it causes concurrency issues when attempting to spawn multiple labels at once
• Possible race condition if two Labels are attempted to be saved at the exact same time to a Classification which doesn't yet exist. The AudioPlayer .addLabel function yields to the creation of the related Classification. If 2 Labels attempt to create the same Classification at the same time within the same transaction frame, they will both see that the Classification as non-existent, try to create it, and one will fail as the unique index on Classification name will not allow it. This will most likely not effect people in real life scenarios, and only came about when doing automated DB seeding.
• electron-forge doesn't play nice with typeorms glob patterns for entity loading as files will not be bundled in the build to import at runtime. Make sure to use the [<Entity>, <Entity>] syntax.
• https://www.tensorflow.org/tutorials/sequences/audio_recognition doesn't work when WAV files are encoded at 44100. Changing to 16000 seems to fix it.

Code Style guide

• .tsx? files which start with a capital do a named class export of the same filename; other files are considered lib files and can export anything.
• Variables suffixed with _ are of type Promise<T> or some unresolved type.

Special Thanks

• This project is heavily influenced by our sister project https://github.com/ritazh/EchoML/