Arousal Valence (msd-musicnn-1.pb; deam-msd-musicnn-2.pb)

[Aradhya-Tripathi]

We are trying too get mood predictions using Essentia models, but we are unsure about the way to use the outputs. We are getting a lists of value pairs, but unsure how to convert / interpret this list of pairs (assuming they are arousal and valence pairs).

[palonso]

Hi @Aradhya-Tripathi,
All our arousal/valence models operate on small chunks of 1 to 3 seconds (depending on your chosen embedding model).
The expected output shape is: (T, D), where T depends on the audio duration, and D represents the valence and arousal values. To predict overall A/V values, you can compute the average along the time axis (T).