Algoriffix has updated Transkr V2, the live and file-based audio-to-MIDI converter, to version 3.0.
Version 3 announces several amendments to the underlying neural network, which all together contribute to a better user experience. These comprise a revised loss function, including regularisers and their scaling, hyperparameter initialisation, and an additional pitch enhancement layer. A spectral emphasis parameter has also been unveiled to give even more control over the model. On top of that, both the multipitch onset detection and the voiced/unvoiced phoneme detection have received a polish.
Mac users can now also obtain a license for Transkr V2 from MuseHub.
