Software Open Access
Complementary source code and data for the publication: Raw Music from Free Movements.
The source code includes the implementation of an adversarial autoencoder for audio waveforms [AAE], a sequence to sequence transducer from sequences of poses to sequences of audio encodings [Seq2Seq]. The implementation is based on Tensorflow v2.3 and has been tested under Ubuntu 20.04.
trained versions of the AAE and Seq2Seq architectures are provided as serialized models and weights.
The data required for testing and training the models is also provided and includes audio files and pre-processed motion capture recordings.