audio_source_classification

Classifying Natural and Electronic Audio

Pytorch Code:

https://github.com/adambielski/siamese-triplet: this has the maximum number of stars. This repo contains both - the triplet as well as the contrastive losses. This repo also has implmenetation of different smaling techniques

https://github.com/zalandoresearch/fashion-mnist: This repo introduces a new dataset: Fashion MNIST

https://github.com/QED0711/audio_analyzer: Visual Audio Analysis --> can be useful for visualizing differences between natural and electronic audio

Useful Code

https://github.com/oscarknagg/voicemap: In Keras, but for speech source classification, embedding space visualization

https://github.com/Rachine/sampling_siamese2018: Sampling Techniques for audio signals code

Relevant papers:

https://arxiv.org/pdf/1706.07567.pdf: Sampling Matters in Deep Embedding Learning: This paper offer an excellent comparision between different sampling techniques and also offers the best performing sampling techniques

https://arxiv.org/pdf/1703.07737.pdf: In Defense of the Triplet Loss for Person Re-Identification This paper goes over using various types of mining methods for triplet loss applied to the task of person re-identification

Blogs:

https://gombru.github.io/2018/05/23/cross_entropy_loss/: This blog goes over different types of losses