Modeling the Compatibility of Stem Tracks to Generate Music Mashups
Authors: Jiawen Huang, Ju-Chiang Wang, Jordan B. L. Smith, Xuchen Song, Yuxuan Wang187-195
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To train a model to predict compatibility, we use stem tracks obtained from the same song as positive examples, and random combinations of stems with key and/or tempo unadjusted as negative examples. ... Finally, we conduct objective and subjective evaluations of the system, comparing them to a standard rule-based system. ... Table 1: Results for objective evaluation. ... Figure 4: Comparison of different systems in subjective evaluation. |
| Researcher Affiliation | Collaboration | Jiawen Huang*,2, Ju-Chiang Wang1, Jordan B. L. Smith1, Xuchen Song1, and Yuxuan Wang1 1 Byte Dance 2 Queen Mary University of London |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | For reproducibility, we suggest one can use Spleeter (Hennequin et al. 2020), an open-source tool with pre-trained models, as a replacement. It is expected to have similar effectiveness for generating the stem samples. ... No concrete access to source code for the methodology described in this paper is provided. |
| Open Datasets | Yes | For testing, we built two Mashup DB s based on two publicly available datasets: Harmonix Set (Nieto et al. 2019) and Ayumix2020. ... The Ayumix2020 dataset originates from the Ayu Creator Challenge event, in which the record label Avex released 100 studio acapella tracks3 from J-pop star Ayumi Hamasaki. ... 3Studio acapella is the clean vocal stem track of a commercially released song. The data are available at https://youtube.com/playlist?list=PL57sdSoJE6THHJyhWfFU7z1RdmWkdmL43 accessed on Mar 10, 2021. |
| Dataset Splits | Yes | We used a ratio of 4: 1 for splitting the in-house dataset into training and validation sets. |
| Hardware Specification | Yes | All models were trained using an NVIDIA Tesla-V100 GPU for 3 days. |
| Software Dependencies | No | The paper mentions software like madmom and Rubberband, and suggests Spleeter as an open-source replacement, but it does not provide specific version numbers for the ancillary software used in their experimental setup (e.g., 'madmom (Bock et al. 2016)' doesn't specify the version they used). |
| Experiment Setup | Yes | All models were trained with Adam Optimization with a 1e 4 learning rate. |