IEVAD - Interactive Embedding Visualizations of Acoustic Datasets

Combine UMAP and interactive Visualizations to explore large acoustic datasets intuitively

ievad uses plotly's dash library to provide an interactive visualization for your acoustic dataset.

UMAP code provided by @avery-b

Best performance of this code is achieved using python version 3.10

Installation

• clone repository

git clone https://github.com/vskode/ievad.git

• create environmen (on windows replace python3.10 with path to python 3.10)
• this step might require installation of virtualenv

python3.10 -m virtualenv env_ievad

• activate environment

source env_ievad/bin/activate

• install depedenciew

pip install -r requirements.txt

• download the model checkpoint, from here
• move model checkpoints from downloads to ./ievad/vggish (see download link above, either manually or using the following code)

mv ~/Downloads/vggish_model.ckpt ievad/vggish

• run program

python run_pipeline.py

Usage

Inside the ievad/files/raw directory is where you can put sound files ending in .wav or .aif and they will then be used for the creation of the embeddings and the visualization of them.

Embeddings can also be computed without visualizing them using python run_embed.py.

Once Embeddings have been created you can just run python run_plot.py to prevent the embeddings from being calculated again.

Edit the ievad/config.yaml file to change the paths to your needs.