Thread by @dietrich_mo

1/ #tweetprint time! how would you like to detect 99% of vocals in recordings automagically? ;)
Then checkout our new preprint on @biorxiv
work lead by @aho_fonseca, with @gumadeiras
Paper: biorxiv.org/cgi/content/sh…
Code: github.com/ahof1704/Vocal…
Dataset: osf.io/bk2uj/

2/ we present VocalMat, an open-source MATLAB-based tool for automatic detection and classification of ultrasonic vocalizations (USVs) from mice 🐭

3/ VocalMat uses adaptive signal and image processing to automatically detect USVs in experimental recordings

4/ using differential geometry and machine learning, VocalMat identifies and eliminates noise reducing the chances for false positives

5/ How about analyzing vocals? VocalMat uses a Convolutional Neural Network to classify USVs into one of 11 categories. This allows for unprecedented detailed analysis of vocals with high-throughput and accuracy!

6/ manifolds! VocalMat uses Diffusion Maps to project vocal class probabilities onto a 3D space for visualization

7/ want to visualize different experimental groups together? you can use the manifold alignment step and visualize embeddings from different recordings in the same space!

8/ we are also making available our dataset with over 10k manually labeled vocalizations!
VocalMat is available on github: github.com/ahof1704/Vocal…
and the dataset is up on OSF: osf.io/bk2uj/

9/ we would love for you to try VocalMat, let us know what you think! dietrich-lab.org/vocalmat

10/ thank you @Yale @IcsnmYale @YaleMed @YaleNeuro @yalepress @ufrgsnoticias @INF_UFRGS @ppgbioq for the institutional support!

11/fin thank you to all funding agencies that have supported the lab and made this work possible!
@bbrfoundation @niddkgov @yalediscovers @choodfoundation @jbpierce_lab @fpwr @CNPq_Oficial @CAPES_Oficial