MARBLE-Benchmark

MARBLE-Benchmark copied to clipboard

How to add a new model?

1

buyaOyiweiniyingle

Hi! I saw that during preprocessing you split examples of vocalset into 3-second chunks and treated them as individual samples when calculating accuracy. I'm wondering why the metrics are not...

sivannavis

Not sure who to reach out to about this, but unfortunately https://marble-bm.shef.ac.uk/ is down due to expired certificate

Laubeee

Add task for MedleyDB dataset

Ningzhi-Wang

Hi! Thank you for releasing this benchmark as an open-source resource, which provides a method for the evaluation of music audio representation models. I want to evaluate the performance of...

XaryLee

GTZAN-genres 复现偏低？

3

你好，使用git中的脚本测试GTZAN-genres的probe结果，mert-v1-95M和330M的结果都偏低，是不是git中的脚本不是论文中得到最终结果的脚本？模型使用的这个https://huggingface.co/m-a-p， 直接测试mert-v1-95M的accuracy 0.7689, 330M的结果0.7379，对比论文的结果0.786，0.793差了比较多。

xujinchang