All models created by the Kits.AI team are exclusively trained on:
fully-licensed proprietary datasets collected by the Kits.AI team with fair compensation to providers
open-license datasets
Open-license datasets used in training include:
Title: Emilia-YODAS dataset
Creator: Amphion (OpenMMLab)
License: Creative Commons Attribution 4.0 International (CC BY 4.0)
Source: https://huggingface.co/datasets/amphion/Emilia-Dataset
Modifications: none
Title: VCTK Corpus (Version 0.92)
Creator: University of Edinburgh, CSTR (Junichi Yamagishi et al.)
License: Creative Commons Attribution 4.0 International (CC BY 4.0)
Source: https://datashare.ed.ac.uk/handle/10283/3443
Modifications: none
Kits also uses the following open source models and software:
Title: Parakeet TDT 0.6B V2 (En)
Creator: NVIDIA
License: Creative Commons Attribution 4.0 International (CC BY 4.0)
Source: https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2
Modifications: none
Title: Audiobox Aesthetics
Creator: Meta AI (Facebook Research); Andros Tjandra et al.
License: Creative Commons Attribution 4.0 International (CC BY 4.0)
Source: https://github.com/facebookresearch/audiobox-aesthetics
Modifications: none
Title: LAME MP3 Encoder (used via lamejs)
Creator: LAME Developers
License: GNU Lesser General Public License (LGPL)
Source: https://lame.sourceforge.io/
Modifications: none
Title: lamejs (JavaScript port of LAME)
Creator: zhuker & contributors
License: GNU Lesser General Public License (LGPL)
Source: https://github.com/zhuker/lamejs
Modifications: none
Title: FFmpeg
Creator: The FFmpeg developers
License: GNU LGPL 2.1-or-later (most of FFmpeg)
Source: https://ffmpeg.org/
Modifications: none
More information on our models can be found at kits.ai/research