Singing voice synthesis for score sonification

TROMPA will generate missing modalities, with a focus on high-quality synthesised (audio) versions of symbolic scores, needed to sonify existing repertoire and return forgotten pieces to life. Singing voice synthesis technology, including choir singing (Blaauw & Bonada, 2017) is available, but the human voice is still one of the most difficult instruments to model. Furthermore, choirs present additional difficulties due to the strong modulations present in the audio, the lack of annotated datasets for training and the fact that choirs sing in several languages.

TROMPA will create multimodal and multilingual digital music repertoire for choirs, developing choir-adapted technologies for automatic transcription by combining knowledge in synthesis and in music description, and creating models of choir voices for synthesis, valid for the different target languages of the project and adapted to different timbres.

 

References

Blaauw, M., Bonada, J., "A Neural Parametric Singing Synthesizer", Interspeech 2017, August 20–24, Stockholm.