Description
microtex is a lightweight version of SCAPES trained on a carefully curated micro-dataset of texture sounds collected from Freesound. Despite its reduced training set, the model retains the core capabilities of SCAPES for texture-oriented sound generation.
A smaller variant of this model, trained on the same dataset, was used in the evaluation presented in the original SCAPES paper, accepted at DAFx 2026.
Data
The training dataset consists exclusively of texture sounds downloaded from Freesound. The table below lists every audio file used for training (with duplicate sources removed), together with the corresponding authors and licenses. This provides full traceability of the training data and proper attribution to the original sound creators.
| Name | Author | License |
|---|---|---|
| Remco_de_Kluizenaar_001_Fire_burning_in_woodstove_sparkles_vuur_kachel.wav | remcodekluizenaar | Attribution NonCommercial 4.0 |
| Applause 27 seconds, Audience c. 200. Large Hall. | iainmccurdy | Attribution 4.0 |
| UH-1 Iroquois Huey - Distant sound and Arrival | SoundFX.studio | Attribution NonCommercial 4.0 |
| Rain_02_wThunder (wav-24bit-96khz) | vincefred | Creative Commons 0 |
| Rain ambient | DRuedaM | Attribution 4.0 |
| BGSaSc Water Beach Ocean Waves Constant Splashing Rocks Many Small Waves Greece 16 | Profispiesser | Creative Commons 0 |
| Screaming Piha in the Amazonian Rain Forest | felix.blume | Creative Commons 0 |
| Small stream in forest | Cinetony | Creative Commons 0 |
| orchestra-tuning-Liceu.wav | xserra | Attribution 4.0 |
| Tokyo-cicadas.wav | xserra | Attribution 4.0 |
| bubbling02.wav | MattJ99 | Attribution 3.0 |
| Saint-Assiscle drainage channel Mono 250323_12h22 | Sadiquecat | Creative Commons 0 |
| keyboard-typing-yellow-switches | aalorv | Attribution 4.0 |