This project aims to construct an open source package to classify vinyl according to the Goldmine Grading.
The package is constructed with my finetuned version of facebook/wav2vec2-base called wav2vec2-base-vinyl_condition
- Obtain grading condition from an audio clip of the vinyl.
Future features:
- Obtain visual grading condition from a photo of the vinyl.
- Obtain classification of pops, clicks, scratches to improve the model accuracy.
- Obtain visual grading condition of the cover from a photo of it.
The main steps of the project included:
- Scraping watchcount to create a database of selled records and links to ebay items.
- Scraping ebay items and obtain audio and and the vinyl condition "raw".
- Labeling the audios from this "raw" conditions.
- Audio exploration and cutting to separate the different sides of the records.
- Preparing the dataset.
- Finetuning the model.
- Evaluation of its performance.
- Creating the python and the gradio applications for production.
The audio and csv files are not uploaded here.