PronounSE Evaluation Dataset for Explosion Sound Synthesis

This dataset provides an evaluation dataset for explosion sound synthesis from language-independent vocal mimicry. The dataset includes vocal mimicry recordings, synthesized sound samples generated from the vocal mimicry using three models, metadata, and a preprocessing script for reconstructing reference sound clips from a commercial sound-effect library. The original commercial reference sound-effect recordings are not included because they are derived from Sonicwire B9642: SFXtools - ST093 - Explosions. Users who wish to reconstruct the reference sounds should purchase the commercial library and run the provided data_preparation.py script. The synthesized samples in Synthesis_Samples/PronounSE, Synthesis_Samples/T-Foley, and Synthesis_Samples/StableAudio2.0 are provided as comparison samples for evaluating vocal-mimicry-based explosion sound synthesis. The Stable Audio 2.0 samples were generated from the vocal mimicry recordings included in this dataset, and no commercial reference sound effects were used as input to Stable Audio 2.0.