MOSA-VPT: a high-quality synthetic violin playing technique dataset

Repository for MOSA-VPT, a novel high-quality synthetic violin playing technique dataset as described in our ICASSP 2026 paper "VioPTT: Violin Technique-Aware Transcription from Synthetic Data Augmentation" (https://arxiv.org/abs/2509.23759v3).

Code accompanying our dataset and paper can be found here: https://github.com/y10ab1/VioPTT

This dataset is derived from the Music mOtion and Semantic Annotation (MOSA) dataset: https://zenodo.org/records/11393449

Dataset contains synethetic .wav files of violin recordings present in MOSA, sampled at 16kHz.

Four violin playing techniques are included (detache, flageolet, pizzicato, spiccato) and divided into 19 zip files per technique for download. 

csv.zip contains alignment metadata of each audio file and midi.zip contains a MIDI transcription of each audio file (both are taken from the original MOSA dataset).