DOSE : Drum One-Shot Extraction

This repository contains the inference script for the DOSE model.

You can view the demo results in the demo section.

You can also check the dataset information.

Drum one-shot extraction task. Given an audio mixture as input, the task aims to output one-shot audio samples for kick, snare, and hi-hat drums. The input audio mixture is encoded into a sequence of discrete tokens using a frozen DAC encoder, which are then fed into a decoder-only Transformer. The Transformer is trained to autoregressively predict the groundtruth drum one-shot tokens by minimizing two losses: onset loss and full-length loss. Finally, the predicted token sequence is decoded into drum one-shot audio using the DAC decoder.

🎧 Check Demo

https://hsuneh.notion.site/DrumSlayer-110005fe1b9443f58668c999c81c5745?pvs=4

How to Run

1. Clone the Repository:

   git clone https://github.com/HSUNEH/DOSE.git
   cd DOSE

2. Install Dependencies

   This code works with Python 3.10.13.

   pip install -r requirements.txt

3. Download Model

   download and locate .ckpt files in ./checkpoints/

   kick.ckpt : https://drive.google.com/file/d/1qgCSwOv43AZNviVKy1rO_FBuuqsj1oGP/view?usp=sharing snare.ckpt : https://drive.google.com/file/d/1FN3Zqvnng_PDsXQ-T6Un87yjU85Gx-4H/view?usp=drive_link hihat.ckpt : https://drive.google.com/file/d/1jtBvDYHDVo8k7gY2_TX_3UucDQ-LJwmP/view?usp=drive_link

4. Evaluation

   python eval.py

   • --inst: Specify the type of drum instrument to extract. Options are kick, snare, or hihat. Example: python eval.py --inst kick

   • --i: Specify the input directory containing the wav files. Example: python eval.py --i ./wavs

   • --o: Specify the output directory where the results will be saved. Example: python eval.py --o ./results

   • --d: Specify the device to use for computation. Options are cpu or cuda:<device_number>. Example: python eval.py --d cuda:0

RMOD (Random Mixture One-shot Dataset)

Dataset generation process. First, kick, snare, and hi-hat loops are synthesized from one-shot drum audio samples using randomly generated MIDI notes. Next, optional bass, piano, guitar, and vocal loops are selected. The drum loops and other musical loops are then processed through independent mixing chains, which apply gain, EQ, compression, panning, limiting, delay, and reverb effects. Finally, all tracks are combined and passed through a mastering chain consisting of EQ and limiter effects.

You can download RMOD from Kaggle.(10000 files)

https://www.kaggle.com/datasets/sunehflower/random-mixture-one-shot-dataset-rmod

Results